热门科技推文 — 2026年8月12日

今日科技领域，人工智能智能体成为焦点：Grok Bot 进入测试阶段，英伟达则发布了 Nemotron 3.5 Lightning 和 NeMo Switchyard，旨在打造速度更快、针对特定任务优化的系统。围绕模型水印技术及所谓提取隐藏推理过程的方法，工程界的争论进一步升温；与此同时，Unsloth 推出了一款用于本地模型训练的开源桌面应用。初创企业的融资势头依然强劲，River AI 宣布完成 11 亿美元融资，据报道 Anthropic 正在为潜在的首次公开募股接洽投资者；此外，IBM 承诺投入 2.4 亿美元，用于建设由英伟达技术驱动的推理基础设施。

1. heyrobinai (Group Score: 1237.7 | Individual: 57.3)

Cluster: 43 tweets | Engagement: 3382 (Avg: 198) | Type: Tech

RT @bot: Introducing Grok Bot, now in early beta.

Bots are AI teammates that do real work for you. They sign in to your tools, use them just like you do, and come back with finished work. https://t.co/uyfA97yo98

See 42 related tweets

@XFreeze: SpaceXAI just launched Grok Bot and this is a much bigger step toward actual AI coworkers

Instead o...

@MatthewBerman: I've been using Grok Bot for the past week.

It's the culmination of where AI agents are going - de...

@lennysan: I got early access to Grok Bot and I'm hooked.

I haven't been this excited about a new AI product i...

@XFreeze: One of the coolest parts of Grok Bot is how customizable it is

You can create completely different ...

@agentnative_: Cursor / SpaceX just released "Grokbot"

Their General Purpose Agent to Rival Claude Cowork and GPT ...

2. FactoryAI (Group Score: 684.8 | Individual: 46.2)

Cluster: 24 tweets | Engagement: 970 (Avg: 143) | Type: Tech

RT @NVIDIAAI: Introducing NVIDIA Nemotron 3.5 Lightning⚡

An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volume, specialized tasks faster.

It delivers up to 4x the output speed of similar-sized models. https://t.co/ENWrZe76pU

See 23 related tweets

@OpenRouter: NVIDIA Nemotron 3.5 Lightning is now live on OpenRouter.

A 30B hybrid MoE with 3B active params, di...

@ctnzr: Nemotron 3.5 Lightning: Same architecture as 3.0 Nano, with added speculative decoding, and with th...
@0xSero: I’ve been testing this model, excellent agent. With DSpark it’s INCREDIBLY fast on a DGX Sparks 240+...
@vllm_project: ⚡ Run Nemotron 3.5 Lightning on vLLM today!

Up to 4× higher throughput and 30% faster task completi...

@kimmonismus: One day after Meta released Muse Glimmer, NVIDIA launched Nemotron 3.5 Lightning and NeMo Switchyard...

3. burkov (Group Score: 567.5 | Individual: 51.6)

Cluster: 27 tweets | Engagement: 2979 (Avg: 195) | Type: Tech

Sounds like bullshit. There's nothing invisible in plain texts. Any plain text editor will show you any "hidden" watermark.\n\nQT @M1Astra: Anthropic says new Claude models will embed invisible watermarks in all generated text, everywhere Claude is offered.

The watermark is part of the text, it isn't metadata: "it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing."

This starts with models launched on or after August 2, 2026, under an EU AI Act code Anthropic signed. Anthropic is still working on adding it to current models. The rollout is worldwide.

See 26 related tweets

@kimmonismus: Anthropic is making invisible watermarks part of Claude at the model level.

Claude models launched ...

@VaibhavSisinty: Anthropic is now watermarking everything Claude writes. Every single output. Globally. 🤯

Every new ...

@testingcatalog: ICYMI 👀: Anthropic will start watermarking AI-generated text via upcoming models released in the EU ...
@jukan05: Anthropic just committed the worst self-inflicted wound possible right before its IPO.\n\nQT @M1Astr...
@Pirat_Nation: Anthropic is working on invisible watermarks for text generated by Claude which would not be visible...

4. negligible_cap (Group Score: 425.4 | Individual: 37.4)

Cluster: 13 tweets | Engagement: 203 (Avg: 108) | Type: Tech

Brad not wasting any time after that tender offer https://t.co/pC8n4NrUUH\n\nQT @bradlightcap: i shared the message below with the openai team this morning. these decisions are never easy, but the talent and mission orientation of the openai team make me more optimistic than ever. 🤍

Team, it is bittersweet to share that I’ll be moving on from OpenAI to start something new. I feel incredibly fortunate to have spent most of the last decade pursuing our mission and building this company. Sitting here today, mission success feels within sight. It has been the honor of my life to help bring us to this point, and to do it alongside all of you.

I joined OpenAI in 2018, eight years ago this month, because I wanted to work on something hard. @sama introduced me to @gdb and @ilyasut, who showed me a pitch deck suggesting that because models improve predictably with scale, AGI was not only possible but likely. If you had told me then that we’d be where we are today, I wouldn't have believed you.

In the years that followed, I had the privilege of building the first versions of most of our operations and business teams – from Finance to Legal, People, CorpSec, GTM/Gov, Partnerships, and more. Among the most rewarding parts of this journey for me has been watching each of these teams mature under brilliant leaders. I am excited for the next decade in the hands of this incredibly capable team.

At the same time, we grew from a small research lab to one of the most consequential companies of our time. I was reminded of that fact constantly – from serving our first user to our billionth, to the growing scale and ambition of our partnerships and the occasional curveball (the Blip, etc.).

Through it all, I’m proud of how we’ve maintained our focus on people. It always amazes me how quickly the world has adopted our tools and rallied behind our mission. I hope we will continue to earn their trust.

Over the last few months, I’ve been focused on the next horizon and what would stand in the way of mission success. I believe there are a few important new things the world will need to get right as we enter this next period. I’ll have more to share soon, but I believe in OpenAI more than ever and am excited to help you all advance the mission from a different vantage point.

I am deeply grateful to have had the opportunity to work with all of you, and to so many of you for the support through the years. The old OpenAI meme that “the real AGI is the friends you made along the way” really rings true for me.

I will be around for the next few weeks. I am not going far and hope to continue to support you all however I can. I’ll always be a phone call away if you need me.

Brad

See 12 related tweets

@AndrewCurran_: Big news. Brad Lightcap is leaving OpenAI to start 'something new.'\n\nQT @bradlightcap: i shared th...
@ns123abc: “Sir… Brad Lightcap… the CFO you hired in 2018 when OpenAI had no product and no revenue… the man wh...
@StockMKTNewz: OpenAI's former Chief Operating Officer, Brad Lightcap just announced he is stepping down from the c...
@KatieMiller: The major leadership positions that have left OpenAI since January:

• Chief Communications Officer ...

@kimmonismus: Brad Lightcap, one of OpenAI’s most important non-technical builders, is leaving after eight years t...

5. nvidianewsroom (Group Score: 301.6 | Individual: 30.7)

Cluster: 14 tweets | Engagement: 304 (Avg: 617) | Type: Tech

Today, NVIDIA announced Nemotron 3.5 Lightning and NeMo Switchyard — two new open technologies for building faster, more efficient AI systems that can use the right model for each task.

Together, they give developers greater control over how and where AI runs.

⬇️\n\nQT @nvidia: Today, NVIDIA announced NVIDIA Nemotron 3.5 Lightning, a customizable model for high-volume, specialized work, and NVIDIA NeMo Switchyard, which helps agents route each workflow step across the models they choose. ⚡ https://t.co/Li96xrOe3K

See 13 related tweets

@StockMKTNewz: NVIDIA $NVDA JUST RELEASED NEMOTRON 3.5 LIGHTNING, ITS MOST EFFICIENT MODEL YET FOR LONG-RUNNING AI ...
@BullTheoryio: NVIDIA LAUNCHES ITS MOST EFFICIENT OPEN AI MODEL

Nvidia has joined the open AI support club, launch...

@TeksEdge: Not enough time or enough local hardware to run @Nvidia's lastest model, OpenRouter has it on tap al...
@wallstengine: Nvidia is developing Nemotron 4, a new open-source AI model expected to have at least 1 trillion par...
@benitoz: NVIDIA's Lightning framing is the tell: agents spend most of their time executing, not planning

The...

6. negligible_cap (Group Score: 246.4 | Individual: 40.7)

Cluster: 10 tweets | Engagement: 265 (Avg: 108) | Type: Tech

ANTHROPIC IS MEETING WITH POTENTIAL INVESTORS TO SHORE UP CONFIDENCE AHEAD OF IPO THAT COULD LAUNCH IN SEPTEMBER OR EARLY OCTOBER - WSJ

Telling investors not to worry is often a great way to make them worry https://t.co/ZzFzpUaeX2

See 9 related tweets

@edzitron: I love to reassure investors and address concerns before my ipo\n\nQT @negligible_cap: ANTHROPIC IS ...
@WSJ: Anthropic is meeting with potential investors to shore up confidence in what could be the largest IP...
@MTSlive: SITUATION DETECTED: Anthropic is meeting with potential investors to shore up confidence ahead of a ...
@zerohedge: Anthropic rushing to IPO (again) while it still benefits from the now-over tokenmaxxing, and before ...
@StockMKTNewz: Looks like we are now roughly 2 months away from the Anthropic IPO

There is now a 74% chance that A...

7. mitsuhiko (Group Score: 225.5 | Individual: 46.5)

Cluster: 10 tweets | Engagement: 1530 (Avg: 147) | Type: Tech

RT @kotekjedi_ml: We can finally talk about it:

We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company.

We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried. https://t.co/S7wN8aP3X7

See 9 related tweets

@teortaxesTex: Amazing. Yeah I've been hearing for a while that this is the real "science of distillation". https:/...
@NielsRogge: https://t.co/Bxf3ZMOvr8\n\nQT @kotekjedi_ml: We can finally talk about it:

We found a way to extrac...

@scaling01: even more reason for frontier labs to just stop releasing small models

not only are they getting co...

@Miles_Brundage: Suspect this won’t get as much attention as the Hugging Face stuff bc it’s harder to explain

But it...

@omarsar0: Recommended reading.

"This suggests that distilling reasoning traces may have been possible for a ...

8. sriramk (Group Score: 223.8 | Individual: 37.0)

Cluster: 8 tweets | Engagement: 164 (Avg: 289) | Type: Tech

really excited for what @ibab is building here.\n\nQT @river_ai_inc: Today, we're sharing that River AI has raised $1.1 billion, led by @generalcatalyst and @amppublic with strategic investment from @nvidia and @AMD. Additional investors include @ycombinator and @Temasek.

We imagine a future where your AI works entirely for you and deeply aligns with your values. Compared to the corporate chatbots of today, it will feel completely different: responsible, curious, and truly yours. Most importantly, you'll own everything: the hardware it runs on, the data it learns from, and the intelligence itself.

Our training API is already live and powering incredible work today. This new funding accelerates our push to bring the full personal AI stack to life.

Read River AI CEO Igor Babuschkin’s (@ibab) conversation with the New York Times to learn more: https://t.co/DK66aKkF31

See 7 related tweets

@AnjneyMidha: Some frontier research leaders retire after initial success

And then there are missionaries like @i...

@garrytan: Deep alignment of your AI with you and your context is massively important

Important new work by @i...

@himanshustwts: River AI has raised $1.1B at$ 5B valuation and they announced it with API Product which is essential...
@WesRoth: this company basically JUST launched and already raised $1.1 BILLION.

River AI was founded by xAI c...

@ycombinator: RT @river_ai_inc: Today, we're sharing that River AI has raised $1.1 billion, led by @generalcatalys...

9. Parul_Gautam7 (Group Score: 195.6 | Individual: 34.6)

Cluster: 7 tweets | Engagement: 140 (Avg: 126) | Type: Tech

1M hours of human video for robotics is a huge milestone, but the real breakthrough is what happens beyond the scale.

Seeing knowledge transfer to unseen robot tasks and different embodiments suggests that human experience could become a powerful pretraining signal for physical AI.

The next big question is how far this can scale and what kinds of human data lead to the best transfer. Exciting direction for general-purpose robotics.\n\nQT @DynaRobotics: Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws:

• world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours, • this human data scaling law implied a scaling law on never seen robot data, • both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge

🧵

See 6 related tweets

@FellMentKE: Dyna-2 proves that physical intelligence scales through human experience.

By training a World-Actio...

@WesRoth: Dyna Robotics just unveiled DYNA-2, a robot foundation model pretrained on more than ONE MILLION hou...
@dr_cintas: Dyna just dropped a world model for robotics, setting a new record for pretraining scale: 1 million ...
@heyrobinai: wait this can't be real

a robot model trained on purely human video is outperforming models trained...

@rohanpaul_ai: RT @rohanpaul_ai: Maybe robots don’t only need more robot data.

Humans already provide a gigantic ...

10. ycombinator (Group Score: 191.7 | Individual: 59.8)

Cluster: 5 tweets | Engagement: 538 (Avg: 44) | Type: Tech

RT @UnslothAI: Introducing Unsloth Desktop 🦥 The first desktop app to run and train models locally.

• Open-source. Runs on Mac, Windows and Linux • Supports MLX, diffusion image/video, audio, GGUF • Connect Claude Code and Codex to local LLMs • 50% more accurate, self-healing tool calls + sandboxed code exec • Works for CPU + multiGPU setups - NVIDIA, AMD, Intel, Mac • Train models 2× faster with 70% less VRAM • Private web search, deep research, RAG, MCP and exports (NVFP4, GGUF) • Use Unsloth’s OpenAI-compatible API and cloud models • Securely deploy LLMs remotely and access anywhere

Unsloth Desktop is now available on https://t.co/xxBDOI9ApK and GitHub.

GitHub: https://t.co/2kXqhhvLsb Blog and Guide: https://t.co/CYosNAQHva

See 4 related tweets

@NVIDIARTXSpark: Training AND running AI models locally from the same desktop app? Yes please. 🔥

Super cool to see @...

@TeksEdge: Oh 💩! @UnslothAI now has a desktop app! 🔥 LM Studio just got some serious competition in the Local ...
@alexocheema: Run and train AI models locally 🦥\n\nQT @UnslothAI: Introducing Unsloth Desktop 🦥 The first desktop ...
@DynamicWebPaige: 🦥 Very proud of this team!!

If you've been looking for a local app to fine-tune open models (on @am...

11. firstadopter (Group Score: 190.6 | Individual: 38.3)

Cluster: 7 tweets | Engagement: 822 (Avg: 134) | Type: Tech

$500 billion will likely go mainly to neoclouds/OpenAI/Anthropic/SpaceX and NOT the top three hyperscalers (Amazon, Microsoft, Google).

Jensen just figured out a way to rapidly level up the smaller guys against the hyperscalers with their ASICs.\n\nQT @firstadopter: Nvidia Scores $500 Billion in Financing for Its Customers. That's A Lot. https://t.co/gvQ4fkqYBZ

See 6 related tweets

@firstadopter: Nvidia may have foreshadowed the hyperscaler hammer back when it reported earnings and changed its s...
@bubbleboi: neoclouds might be what memory was in the first half of the year this is so bullish\n\nQT @firstadop...
@MilkRoadAI: RT @WhiteCollarExit: 3 bullish implications of the $500B$ NVDA consortium (Save this)

First, a sixt...

@firstadopter: Following the Nvidia $500 billion financing news.

Coreweave and Nebius UP, Amazon/Google/Microsoft...

@firstadopter: Hyperscalers weak after the market open. Ruh roh.\n\nQT @firstadopter: $500 billion will likely go m...

12. Parul_Gautam7 (Group Score: 186.9 | Individual: 33.9)

Cluster: 7 tweets | Engagement: 65 (Avg: 126) | Type: Tech

The interesting shift is from buying generic AI to building intelligence that gets better with your own data, workflows, and experience.

That’s where the long-term advantage could come from.\n\nQT @Koukoumidis: Enterprise AI is in a wildly paradoxical state 🤔, and we’re reaching the inflection point that will resolve it. 💥

Enterprises want to differentiate with AI, yet rent the same intelligence as their competitors.

Their workflows and expertise are highly specialized, yet they rely on generic models built to be good at everything.

They worry about AI costs, yet pay premium prices for massive models where only a fraction (1%) of the intelligence is relevant to their task. 💸

And they demand control and sovereignty, yet rent the intelligence becoming core to their business.

This is not a sustainable equilibrium.

The next era of enterprise AI is specialized intelligence companies build, own, and compound. And we are at the inflection point of this transition.

That’s the bet we made when we started @oumi_ai two years ago.

Today we’re closing the loop: Oumi can now not only automatically build your specialized AI models, but also deploy them into production, learn from their production experience, and continuously improve them.

The intelligence that your business runs on, becomes your differentiator. Your compounding advantage.

The winners of the next AI era will turn their own data, expertise, and experience into specialized intelligence that nobody else can rent.

Don’t rent your AI. Build it. Own it. Compound it.

See 6 related tweets

@TheoBuildsAI: RT @Koukoumidis: Enterprise AI is in a wildly paradoxical state 🤔, and we’re reaching the inflection...
@TheoBuildsAI: The feedback loop is the piece I’d watch closely here.

Most AI products generate an enormous amount...

@Origin_AI_01: The biggest mistake in enterprise AI may be confusing access with ownership.

If your competitors ca...

@Rajananandan: If you are the CEO or CIO of a large enterprise, this one is for you...\n\nQT @peakxvpartners: The r...
@Parul_Gautam7: RT @Parul_Gautam7: The interesting shift is from buying generic AI to building intelligence that get...

13. TuhinChakr (Group Score: 174.3 | Individual: 57.5)

Cluster: 5 tweets | Engagement: 930 (Avg: 43) | Type: Tech

RT @alexcdot: Claude's watermark probably doesn't work how you think. As the CTO of GPTZero, I'll explain how Anthropic, Google and OpenAI are building text watermarking in this brief explainer and whether it can be defeated.

Almost all forms of watermarking that are fast and cheap enough for a frontier lab have the same formula, following the KGW method:

In generation:

Let's say you've generated n tokens so far. Take those n tokens + a secret key to generate a random hash
Use that hash to randomly reweight the probabilities for the n+1 token, and then sample from that new distribution. In the simple case, you could split 50% of all English words into a green or red set based on your hash, and boost the probability of words in the green set.

For watermark detection:

For each token, see if it was in the green or red set.
To do this, recreate the hash based on the secret key and the text preceding the current token. Then, recreate the green and red set of words.
Once you've checked all the words in the text, if the next token is selected disproportionally from the green set more than 50% of the time, you claim the text has the watermark.

I can tell you want to ask the following:

Isn't it easy to mess up the hash if you paraphrase the text? The answer is mostly yes, however, you can use a statistical model to get your hash instead of a deterministic function (SIR, Adaptive Watermark). Since the entire watermark is probabilistic, this is fine.
Doesn't this make the text much worse? The answer is yes, it does - Yes, it does – but for most people, it's imperceptible (Google claims in human feedback study with 20,000 texts), since there are exponentially many ways to write the same paragraph. DiPmark does something more sophisticated to avoid shifting the text distribution on average. Of course, watermarks fail on short text or highly predictable texts like "2+2=4".
Shouldn't it be easy to figure out the green and red sets? The answer is no. You would need an exponentially large number of samples from the watermarker to reconstruct those sets exactly, but it's a risk if the detector is open to the wild (Watermark Stealing)

Still, there are couple challenges that a frontier lab needs to overcome:

Their watermark needs to work token-by-token because they are streaming their text to users. Many watermark methods plan sentences or paragraphs at a time, or change the text after its entirely written, in order to make their watermark robust to paraphrasers, and a frontier lab cannot afford to do this yet (SemStamp, PostMark)
If the secret key leaks, the watermark is busted. To avoid a large blast damage from this, you need to have a couple secret keys in rotation.
There are some texts, like code, that cannot be arbitrarily changed, otherwise the code will break. In those cases, the watermark needs to selectively change words in parts of the text that can tolerate synonyms (i.e. like variable naming) - see SWEET, EWD, Invisible Entropy.
They will need to educate their users on how to deal with false positives and false negatives of a detector, which is a big challenge (one we put a lot of effort into)

So, how do I see this playing out in the next 6 months?

If Anthropic releases the watermark detector publically, I think they defeat their own watermark. People find reliable watermark removal strategies by testing against Anthropic (AI detectors like GPTZero have an advantage here because they can train against these adversaries once they become popular).
If they keep the detector private to the government, like Google has done, it's "safer". However, there are some papers showing trained approaches that work robustly to zero-shot break watermarks without any data, simply because they try to write the text just like a human (Zhang et al. 2024, Watermarks in the Sand). Also, making your detector makes it battle-tested and stronger long-term (my experience).
In my testing, the watermarks don't survive intense paraphrasing (especially if you combine word choice and syntax attacks), or human text substitution (rewrite your AI text by plagiarizing human authors). The free paraphrasers I've tried have quickly bypassed Google Deepmind's SynthId for what it's worth.
All-in-all, frontier labs are likely okay with this because they expect most users to not attack the watermark, and also because they + European regulators likely don't care past a certain point - its good enough.
Overall, I think users of frontier LLMs will not really care about this, because 1) they don't realize watermarks are there, 2) EU will force everyone to conform, 3) this seems more like regulatory hoop-jumping than an earnest effort from frontier labs to expose LLM use

Lastly, people's first concern shouldn't be watermarking, it should be AI detectors!

If you're posting, "its not X, its Y!!", I don't think the watermark is going to make a difference :)

See 4 related tweets

@Dan_Jeffries1: You know where altering text is not just imperceptible and a "minor" nuisance, to say nothing of the...
@mark_k: Highly informative post explaining the inner workings, upsides, and downsides of text watermarking, ...
@TaylorLorenz: “Doesn't this make the text much worse? The answer is yes, it does - Yes, it does”\n\nQT @alexcdot: ...
@Michaelzsguo: Anthropic is reportedly planning to watermark text generated by Claude. That sounds alarming, but it...

14. cryptopunk7213 (Group Score: 166.8 | Individual: 37.9)

Cluster: 5 tweets | Engagement: 332 (Avg: 78) | Type: Tech

honestly a genius move from jensen. banks are going to eat this shit up

pitching GPUs as a productive asset that backs the global economy cements infinite capital and demand for nvidia’s products.

and tbh they have a distinct advantage over competitors:

nvidia gpus are broad = fungible. can be used for llms, video/image models, inference or training, fine tuning etc. any workload.
CUDA moat glues every customer to the same software stack. also optimizes exiting hardware to run efficiently.
depreciation cycle of their gpus are the longest. these things go for 10+ years.

as jensen puts it - all hallmark traits of an productive asset.\n\nQT @JensenHuang: https://t.co/pppEbPVtfE

See 4 related tweets

@VaibhavSisinty: Nvidia built the GPU that runs every AI model you use. Now it is building the financial system to fu...
@WesRoth: Frontier labs and AI clouds may have enormous demand for GPUs but building multi-gigawatt AI factori...
@VaibhavSisinty: Jensen Huang went live with Goldman Sachs, BlackRock, Blackstone, KKR, Brookfield, and Apollo. All o...
@coinbureau: 🎥WATCH: Nvidia’s Jensen Huang alongside CEOs of Goldman Sachs, BlackRock, Blackstone, KKR, Brookfiel...

15. StockMKTNewz (Group Score: 163.3 | Individual: 35.5)

Cluster: 8 tweets | Engagement: 243 (Avg: 205) | Type: Tech

IBM $IBM JUST SIGNED A$ 240M MULTI-YEAR DEAL WITH TOGETHER AI TO DEPLOY THE FIRST LARGE-SCALE INFERENCE CLUSTER OF ITS KIND ON IBM CLOUD

The cluster runs on NVIDIA $NVDA HGX B300 systems and Spectrum-X Ethernet networking, expected to be available in Q1 2027. It's the first dedicated, large-scale cluster built for inference on IBM Cloud using this hardware, and NVIDIA says it's built to deliver 30x more AI factory output than prior generations.

Together AI will use it to run inference for open-source models. The company recently raised an $800M Series C round at an$ 8.3 billion valuation and says it's now serving 400 trillion tokens monthly.

See 7 related tweets

@StockSavvyShay: $IBM signs a$ 240M multi-year deal with Together AI to deploy a large-scale inference cluster on IBM...
@wallstengine: $IBM and Together AI signed a multi-year$ 240M deal to deploy a large-scale AI inference cluster on ...
@Polymarket: JUST IN: IBM & Together AI sign a $240 million deal to build a large-scale Nvidia-powered AI inf...
@WOLF_Financial: All roads in AI lead back to Jensen

There is currently a 70% chance that Nvidia $NVDA ends 2026 as ...

@Techmeme: IBM and Together AI sign a $240M, multi-year deal to build an AI inference cluster on IBM Cloud, usi...

16. XFreeze (Group Score: 156.9 | Individual: 36.0)

Cluster: 5 tweets | Engagement: 569 (Avg: 570) | Type: Tech

Grok Build v1.0.1 just landed with a serious upgrade, bringing safer multi-agent scaling, smarter session management, expanded video generation, stronger reliability, and major performance gains for large-scale development

Release Notes: v1.0.1

Breaking Changes: • /rewind now only truncates conversation history instead of files as well and asks for confirmation by default. • Managed MCP servers are now only available through the gateway catalog.

Features: • Subagent spawning is now bounded; wide fan-outs queue instead of exhausting file descriptors. • New grok du command shows disk usage of ~/.grok including worktrees and sessions. • Tools now report whether they only read data, enabling safer restricted agents and subagents. • Sandbox workspace sessions can now limit which bundled skills are advertised via caller config. • Renaming a session from the dashboard now starts with the current title prefilled for easy editing. • /usage, /session-info, and /context now open in a tabbed modal instead of adding text to the conversation. • grok trace exports now bundle memory trace files for easier debugging. • Session rename now enforces a 100-character limit, ghost-prefills the current title, and preserves manual titles across machines. • New /rename --auto command unpins a manual session title so automatic titling resumes. • Video generation from references now supports preset voices, single-image input, 1–15 s durations, and 4:3 / 3:4 aspect ratios.

Bug Fixes: • Sandbox config entries ending in /** now correctly grant the parent directory instead of creating a literal ** subdirectory. • Failed alpha/enterprise updates now suggest the matching GROK_CHANNEL reinstall command. • On Apple Silicon, grok now installs the native arm64 build even from a Rosetta shell or x86_64 updater. • Skills that share names with built-in commands now appear alongside them in the slash menu with qualified names. • Notebook permission rules imported from Claude configs are now ignored with a warning instead of applying broadly. • Goal evaluation at round end no longer fails due to timeouts. • Tool timeouts no longer cause the agent to hang when child processes are stuck in D-state or hold pipes open. • Home and End keys now move to the start or end of the current logical line even when the prompt is wrapped. • Worktree sessions now correctly show their branch in the status bar. • Worktree sessions now keep their status correctly when switching directories or resuming. • Worktree status is no longer lost when opening the dashboard. • Recaps are now written in the same language as your conversation. • Plugin suggestions no longer flash incorrectly while typing. • Send Now now works during active goals without cancelling the goal. • Headless sessions now correctly wait for MCP servers when using delivery tools. • Non-interactive sessions (grok -p) no longer fail when the agent asks for user input or plan approval. • read_file errors for missing skills now suggest the correct registered path instead of a generic hint. • Session load and creation can no longer freeze forever when .envrc evaluation blocks. • Upgraded installs no longer silently run outdated platform skill instructions. • Deleting a session now properly stops and waits for any running subagents before wiping history. • Permission and plan-approval notification hooks no longer fire on auto-allowed tools. • Mid-turn steering sent with double-Enter or Ctrl+Enter now correctly tells the model it arrived while work was in progress. • Scrollback drag selection no longer gets stuck after the mouse button is released outside VS Code or Cursor terminals. • Video generation tools now show a clear error explaining ZDR storage requirements instead of silently disappearing. • Esc on the cancel-turn panel now closes the panel and keeps the current turn running as the shortcuts bar indicates.

Performance: • Git status and diff operations no longer cause high CPU or memory use on large repositories. • Large git histories no longer cause excessive memory use or unresponsiveness. • History search no longer leaks background threads in long sessions with many subagents. • Resuming large sessions is now significantly faster and the UI no longer shows an incomplete transcript while replay is still applying.\n\nQT @XFreeze: Grok Build officially reaches v1.0.0, rapidly becoming one of the most powerful coding harnesses in the world

After 100+ updates in just over 10 weeks, SpaceXAI’s coding agent has hit a major milestone.......with a smarter dashboard, a more polished developer experience, stronger reliability, safer permission handling, and meaningful performance improvements across the platform

The pace at which SpaceXAI is shipping Grok Build is totally insane

Release Notes: v1.0.0

Features: • Dashboard rows show a short summary of what the agent did in the previous turn • Extensions modal groups items alphabetically with collapsible Skills sections • Grok skips the project-directory prompt when launched from home or other non-project directories • /feedback opens a dedicated report box instead of prompt mode • Auto theme detection works over SSH and inside tmux • Markdown tables reflow inside cells on narrow panes instead of clipping • Permission prompts show the complete script; long bash bodies expand with Ctrl-F

Bug Fixes: • MCP tools that return images no longer drop or corrupt large screenshots • Sandboxed Grok starts on large directories with many deny-glob matches • Rapid send-now presses no longer lose earlier queued messages • Esc and stop prevent background tasks from restarting the model after cancel • Login no longer skips when an invalid API key is in the environment • Model picker and command palette work while reviewing a plan • Tab and Esc behave consistently on question, permission, and cancel-turn cards • /new from the dashboard returns to the dashboard from an empty prompt • Codebase restore no longer hangs on large or shallow git repositories • Remote resume restores conversation only unless --restore-code is passed • Copying CJK text with the mouse includes every character at the selection edges • API errors appear as clean banners instead of raw JSON dumps • Typing exit or quit in the dashboard exits the CLI • Mode indicator (plan/agent/ask) stays in sync after resume and mode changes • /delete returns to the dashboard when you delete a session opened from it • Enter in the slash command menu runs the highlighted command • Grok retries more server errors during outages • Session-only slash commands show a message when used from the dashboard • Queued prompts stay visible while waiting on subagents, and slash/image rows can be reordered • Auto recaps no longer appear mid-turn or while busy • /btw error messages wrap fully

Performance: • Forking very large sessions no longer uses many times the session file size in memory • Exiting an empty session is instant, even on slow networks

See 4 related tweets

@elonmusk: What would you most like us to add to or fix about Grok Build?\n\nQT @cb_doge: 🚨 NEW GROK BUILD UPDA...
@cb_doge: 🚨 NEW GROK BUILD UPDATE 🚨

v1.0.1 — 2026-08-10

Updates: • /rewind now only truncates conversation ...

@blankspeaker: SpaceXAI: Grok build now shows more details when you look up /usage , you have options of looking ...
@blankspeaker: Grok Build v1.0.1 is out with a few practical updates. Standouts include /rewind now only truncates ...

17. pcuenq (Group Score: 155.5 | Individual: 45.7)

Cluster: 5 tweets | Engagement: 2056 (Avg: 291) | Type: Tech

RT @AIatMeta: Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows.

Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on consumer hardware like a Mac or PCs with performant GPUs.

In keeping with our long tradition of sharing fundamental AI research, we’re releasing model weights under a permissive Apache 2.0 license.

🧵👇

See 4 related tweets

@Qualcomm: .@AIatMeta's open-weight Muse Glimmer model is optimized for @Snapdragon powered devices, bringing i...
@baseten: Muse Glimmer, Meta's new open-weight agentic model, is available on Baseten day 0!
30B parameters...
@quxiaoyin: Damn Meta. You are killing so much margin\n\nQT @AIatMeta: Introducing Muse Glimmer, an open-weight ...
@pstAsiatech: 👇👇👇\n\nQT @AIatMeta: Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for loca...

18. Dimillian (Group Score: 154.6 | Individual: 47.2)

Cluster: 7 tweets | Engagement: 946 (Avg: 92) | Type: Tech

RT @OpenAI: Now in preview: The ChatGPT desktop app for Linux.

Use ChatGPT, ChatGPT Work, and Codex where you already work and build, with your projects and browser workflows on supported Linux systems. https://t.co/OtsPt5N5QC

See 6 related tweets

@btibor91: The Codex/ChatGPT desktop app is finally available on Linux, too

I've been testing it for a while o...

@thsottiaux: We did it, finally... Codex & ChatGPT desktop, now on Linux.

Thanks for waiting and you can ca...

@dkundel: Here's a picture of a baby penguin to celebrate https://t.co/RbyJ6et7oG\n\nQT @OpenAI: Now in previe...
@TheRealAdamG: https://t.co/jYAKow25vO

Codex App for Linux\n\nQT @OpenAI: Now in preview: The ChatGPT desktop app ...

@PaulSolt: Use Codex on LINUX.\n\nQT @OpenAIDevs: Build where your code already lives.

Codex is coming to Linu...

19. heyshrutimishra (Group Score: 140.2 | Individual: 30.7)

Cluster: 5 tweets | Engagement: 144 (Avg: 29) | Type: Tech

RT @OhansEmmanuel: Introducing Coldtea.

The ADE for shipping self-driving software. Purpose-built for the entire software lifecycle.

Other IDEs optimise for one thing: Build. Run more agents in parallel, ship faster, orchestrate better.

But software engineering is not just about building.

@ColdteaAI brings your coding agents, agentic end-to-end testing, and production monitoring agents into ONE development environment, so you can move at agent speed without breaking anything.

See 4 related tweets

@Parul_Gautam7: Coldtea is tackling the part of AI coding that often gets overlooked: what happens after the code is...
@cgtwts: Vibecoders when they realise Coldtea can have agents test what they build, find what breaks, and fix...
@svpino: Most AI-generated code is slop that goes nowhere.

Anyone can write code, but building software that...

@heynavtoor: A development environment that assumes agents will handle coding, verification, and observation.

Th...

20. HarryStebbings (Group Score: 138.7 | Individual: 36.5)

Cluster: 4 tweets | Engagement: 150 (Avg: 123) | Type: Tech

To win in AI, build a specialised model

"No one's gonna win all of it. You're not gonna build a model that wins all of it.

You might as well build a model that is known to specialize in something very useful and that's very important to your company and your business.

A lot of enterprises are gonna move that direction, make their own models, make their own branded intelligence." @alexatallah

How do you think about this and what does no one know about specialised models that everyone should know @winstonweinberg @lqiao @ceo_clickhouse\n\nQT @HarryStebbings: OpenRouter is one of the most insane stories in tech.

Co-founded by OpenSea founder, Alex Atallah. It has become one of the most important companies in AI.

Scaled to 100s of millions in revenue and wildly profitable.

They process 25 TRILLION tokens every week and will do 1 QUADRILLION tokens this year.

As a result, Stripe have reportedly offered to buy them for $10BN.

Last round was $1.3BN just months ago… so what happens now?

I sat down for a chat with OpenRouter Founder, @alexatallah and have summarized my notes below:

This Is Going to Be the Biggest Market in the History of Tech

AI model inference is set to become one of the largest markets in technology. Even as token prices fall dramatically, Jevons Paradox means usage can grow far faster than costs decline, driving massive overall compute consumption across an increasingly multi-model future.

Why None of the Routing Products Being Created Today Will Compete With OpenRouter

Building AI gateways has become trendy, but copycat routers that treat routing as a side feature are playing to exist rather than to win. True routing requires relentless focus on optimizing latency, cost, and constantly shifting model quality while giving developers maximum flexibility.

Model Labs Have Every Incentive to Come After Your Startup Eventually

Startups building thin wrappers around AI models face an existential threat if they occupy workflows that frontier labs consider strategic. Releases like Claude Design show how labs can move up the application layer, absorb valuable use cases, and lock entire enterprise teams into their ecosystems.

One New Model Every 10 Hours and the Myth of Model Consolidation

Model creation is accelerating as hardware companies, Neo Labs, and agent frameworks continuously release specialized models. Rather than consolidating around a single winner, developers are embracing diverse models optimized for different tasks, reinforcing a fragmented, multi-model future.

America Is Still Very, Very Behind in Open-Weight AI

The U.S. remains significantly behind China in the race for open-weight models. Chinese labs benefit from aggressive state support and open-source momentum, while American efforts face business model and capital constraints. Closing the gap will require dedicated access to compute and sustained investment.

How Open-Weight Models Are Rapidly Closing the Gap on Frontier AI

Models like Kimi K3 and the 5.2 generation show how quickly open-weight intelligence is closing the gap with proprietary frontier models. Using low-cost open models for deterministic subtasks beneath a frontier orchestrator can maintain output quality while sharply reducing inference costs.

Why We Need to Think About Employee Cost Completely Differently

In the AI era, employee cost is no longer just a fixed salary. It increasingly includes the variable inference spend of the AI tools each employee deploys. Companies will need to evaluate workforce performance through a cost-to-productivity lens that accounts for both human compensation and AI consumption.

(links in comments)

See 3 related tweets

@amitisinvesting: A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY.

Here's a full recap:

Nvidia $NVDA announced...

@FellMentKE: Renting a broad model forever can help with access, but it does not create much memory for the busin...
@ai_explorer25: We spent a whole quarter tuning prompts to push a general model into one specialized job. A purpose-...