热门科技推文 — 2026年8月26日

今日科技界，人工智能硬件与智能体成为焦点。苹果推出面向本地模型运行的 M6 Mac Mini，而据报道，OpenAI 的“Jalapeño”推理芯片展现出颇具潜力的通用性能。最新基准测试显示，编程智能体在全代码库重构方面仍面临挑战；与此同时，开源智能体系统以及能力日益增强的机器人和视频模型持续取得进展。初创企业热潮延续，Icon 完成 3000 万美元融资，Anthropic 发布了令人瞩目的营收预期；SpaceX 则宣布计划在路易斯安那州建设一座投资达 1000 亿美元的太空港。

1. SawyerMerritt (Group Score: 594.3 | Individual: 41.4)

Cluster: 19 tweets | Engagement: 7809 (Avg: 2016) | Type: Tech

SpaceX has just officially announced Starbase, Louisiana, a new $100+ billion spaceport in South Louisiana near Pecan Island.

Construction starts in 2027, with the first Starship launch from the new site planned for 2029.

Starbase, Louisiana will be built to support thousands of Starship flights a year. This self-sustaining spaceport will have propellant production, power generation, deep-water shipping capabilities, vehicle processing facilities, and an airport.

SpaceX is committed to investing at least $100 billion and creating more than 3,000 new jobs at Starbase, Louisana hiring heavily from nearby communities and utilizing Louisiana workers’ extensive experience of building complex infrastructure in their unique environment.

In Vermilion Parish, the shoreline is eroding between 3.3 and 23 feet per year. SpaceX is partnering with state and federal agencies to expand Louisiana’s Coastal Master Plan and Coastal Wetlands Planning, Protection and Restoration Act projects, including Gulf shoreline protection breakwaters designed to reduce wave energy and slow loss along the Gulf edge.

See 18 related tweets

@elonmusk: All accurate\n\nQT @SawyerMerritt: SpaceX has just officially announced Starbase, Louisiana, a new $...
@cb_doge: BREAKING: SpaceX has just announced a massive $100 BILLION Starship launch site in Louisiana, the la...
@MTSlive: SITUATION EXPLAINED: SpaceX is building a $100 billion spaceport in Louisiana.

• It goes on roughly...

@pbeisel: Starbase Louisiana

Perfect site. Bigger than Starbase Texas.

• Plans call for five launch complexe...

@WOLF_Financial: $SPCX TO BUILD NEW STARSHIP LAUNCH SITE IN LOUISIANA

SpaceX announced plans for “Starbase, Louisian...

2. Scobleizer (Group Score: 518.8 | Individual: 32.0)

Cluster: 26 tweets | Engagement: 156 (Avg: 114) | Type: Tech

RT @VaibhavSisinty: Apple is no longer hiding it. These machines are built for local AI.

Apple unveiled the new Mac Mini with M6 their first 2nm chip. Starting at $899.

What you can run on the $899 Mac Mini (32GB) 👇

→ Qwen 3.6 27B : fits comfortably, near-GPT-4 quality → Qwen 3 Coder 32B : full coding assistant on your desk → Llama 4 8B at full precision → DeepSeek R1 32B for reasoning tasks → 13.5x faster LLM processing than M1. 4x faster than M4.

What you can run on the Mac Studio with M5 Ultra (512GB) 👇

→ Llama 4 70B at high quality quantization → DeepSeek V4 with hundreds of billions of parameters → Frontier-scale models entirely in local memory → Connect four Mac Studios over Thunderbolt 5 cluster them to run trillion-parameter models locally

Apple said their Mac business grew 30% last quarter largely because people are buying these to run AI models on their own desks instead of paying for cloud tokens.

The future of inference is local. Apple built the hardware for it.

See 25 related tweets

@minchoi: Apple just dropped the M6 Mac mini.

4x faster AI. Up to 4.8x faster local LLMs vs M4. Built for alw...

@heyshrutimishra: Apple quietly announced the most important Mac ever.

No keynote. No event. Just a press release at ...

@wallstengine: $AAPL LAUNCHES NEW MAC MINI WITH M6 AND M5 PRO

Apple unveiled a new Mac mini lineup, with preorders...

@IntCyberDigest: Apple is leading the local AI revolution, and seriously undercutting NVIDIA along the way.

Mac Stud...

@rohanpaul_ai: Massive move by Apple for local inference. They just launched M6 and M5 Ultra to move more AI infere...

3. TheoBuildsAI (Group Score: 467.7 | Individual: 39.7)

Cluster: 14 tweets | Engagement: 144 (Avg: 49) | Type: Tech

RT @EinsiaAI: The ultimate test for coding agents isn't local editing— it's whole-repo evolution, and right now, the survival rate is 5.4%.

Today we’re releasing SWE Refactor Bench, a benchmark for long-horizon, whole-repository software stack migration.

Coding agents are getting very good at fixing bugs. But can they refactor an entire system, C → Rust, Maven → Gradle, POSIX → WebAssembly?

We built 20 real migrations across projects, including SQLite, zlib, libsodium, and GraphHopper.

520 runs. Only 28 survived all 3 stages. 13/20 tasks were solved by nobody.

System-scale migration is still wide open.

Full breakdown 👇 GitHub: [https://t.co/xXyLQ3qq0C] Paper Link: [https://t.co/lO1Enh63q7] Einsia Website: [https://t.co/AGn0hF5gwL]

See 13 related tweets

@TheoBuildsAI: C → Rust sounds simple when you write it in one line.

Doing it across a real repository while keepi...

@NovaIAHQ: 47/100 for the best coding agent on real repository migrations.

That’s the number that caught my at...

@EmmaUsesAi: SQLite, zlib, libsodium, GraphHopper.

These aren’t artificial coding puzzles built specifically for...

@Parul_Gautam7: The adversarial validation result is especially telling.

Even among submissions that passed the fix...

@alexaiworks: A coding agent can pass thousands of tests and still fail the actual job.

That might be the most im...

4. TheoBuildsAI (Group Score: 355.9 | Individual: 34.6)

Cluster: 11 tweets | Engagement: 37 (Avg: 49) | Type: Tech

this guy just posts what actually works in production AI:\n\nQT @mardehaym: A PE-backed healthcare company asked us to modernize their prior authorization automation on a ten-year platform with regulated patient data.

The product roadmap called for a full platform rebuild with a new architecture, new data model, and new automation engine.

They were skeptical of another automation overhaul. We proposed a proof of concept: one live payer portal, prove it works before building the product.

We shipped a production browser extension in two months. Coordinators stopped bypassing the system.

The bots ran headless browsers from a data center. Payers detected the IP and flagged it. When a bot hit MFA or a clinical question it couldn't answer, the session died.

The coordinator filed a support ticket, waited for engineering to patch that specific workflow branch, and started the submission from scratch.

Payer portals branch depending on procedure codes, clinical context, and plan requirements. The old system learned those paths one failure at a time.

A coordinator would submit, the bot would break on a new branch, engineering would patch, and the next case would hit another branch nobody had mapped. The portal backlog never closed.

We moved the automation into the coordinator's own browser. Same authenticated session they'd use for a manual submission. Payers see a normal login from a normal IP. When the automation needs help, the coordinator steps in and the run continues where it left off.

Concept to production in two months with six engineers, HIPAA-hardened, PHI remediation across stored values and logs.

Then we built Observe Mode. When a coordinator works a portal that doesn't have a script, the extension captures what they do. Those recordings become the foundation for the next automation.

Coverage that builds from real workflows instead of breaking first and patching after.

Ten years of payer workflow knowledge baked into the scripts. We migrated it into the new engine while the legacy platform kept running underneath.

Build beside it, prove it works, and let the old system go when it's ready.

That's how we modernize platforms that can't go down.

See 10 related tweets

@TheoBuildsAI: RT @mardehaym: A PE-backed healthcare company asked us to modernize their prior authorization automa...
@TheAva_AI: this guy posts better AI case studies than most consulting firms charge for.\n\nQT @mardehaym: A PE-...
@Lacoste0x_: the best AI account for anyone in PE or engineering leadership\n\nQT @mardehaym: A PE-backed healthc...
@OliviaReedai: best follow on X if you care about AI in production:\n\nQT @mardehaym: A PE-backed healthcare compan...
@NovaIAHQ: the only AI engineering account that posts real numbers:\n\nQT @mardehaym: A PE-backed healthcare co...

5. arsh_goyal (Group Score: 350.7 | Individual: 62.5)

Cluster: 10 tweets | Engagement: 732 (Avg: 58) | Type: Tech

RT @icon: 3 months ago, I sold my 1st company, Skio, for $105M cash.

Now my 2nd company, Icon, just raised $30M to launch something the world has never seen.

For $1000, we make 6 human creator ads (no AI, 100% real). We find creators, ship products, write scripts, & edit videos.

It's risk-free:

Full refund if you don't love your ads.
You don't pay until you review 6 scripts.
Unlimited human revisions.

Today, Icon introduces the world's first Agency.

We're not building a normal agency:

We bought a $12M domain.
We're backed by Founders Fund and OpenAI + Google DeepMind leaders.

It’s impossible to hire a creative team like ours for even $30K/mo:

Creative Directors who've worked with $100M brands (IM8, Obvi, Ridge). They update our format library daily based on winning ads (street interviews, skits, & 20+ more).
12+ scriptwriters with $80M in Meta adspend.
200+ in-house editors who've edited 100K ads.

One more thing...

We're giving away our internal Google Drive with 1000 winning ads that have driven $100M+ in revenue.

You can make $100K copying these ads if you know what you're doing.

Want the Google Drive?

Retweet this post.
Comment "icon".
DM me this post’s link & I'll DM you back with the Google Drive.

[Bonus] Get 3 free human creator ads if you do the above and we end up working together.

See 9 related tweets

@Meer_AIIT: this might be the most important pivot in AI right now and barely anyone noticed.

Icon launched in ...

@StockMKTNewz: Icon just announced it has raised $30M

“For $1000, we make 6 human creator ads (no AI, 100% real). ...

@shiri_shh: > be Kennan Frost

drop out of college at 19 fail through YC pivot into Skio hit $10M ARR in ...

@itsolelehmann: tbh i'm a massive AI UGC hater.

i literally think the whole thing is fraud

the point of UGC is tha...

@VadimStrizheus: 6 ads can save you 11 months of building the wrong product 😭

the entire test costs $1,999.

id run ...

6. Scobleizer (Group Score: 331.1 | Individual: 34.0)

Cluster: 11 tweets | Engagement: 61 (Avg: 114) | Type: Tech

I ran the test again this morning.

Asked an LLM about a company I have been covering for years. Not wrong. Just average.

Like it read the homepage once in 2023.

Andrey ran Yandex Search against Google. Then Amazon AGI. Now he built his own index so agents can query the live web.

Most AI search is still a wrapper. This crawled it. Accel and Conviction put 26 million in. API is free through September.

@styskin\n\nQT @styskin: Today we are announcing @KeenableAI, an AI-native index of the best human knowledge we have, starting with the open web.

AI seems to know just about everything until you ask it about something you know deeply. The answer isn’t wrong, but it's just very average.

We started Keenable to fix this problem: every model and every agent should be able to query, reason over, and continuously learn from the living web.

Backed by a $26M Seed from @Accel and @conviction. Built by the team that took on Google at Yandex Search, with researchers and engineers from Amazon AGI, X, and Perplexity.

The Web Search API and Web Query Language for AI are live now. Together they make the web queryable at AI scale. Both are free until end of September.

Seven days of benchmarks, stories, and launches ahead.

Let your agents search.

See 10 related tweets

@conviction: let the agents search!

@pranavreddy backing @styskin (who ran Yandex web services and search) + te...

@aakashgupta: Keenable stepped out of stealth today with a seed round. Founder Andrey Styskin is on his third sear...
@svpino: These guys are building a new search engine for agents:

• They have their own crawler • They have t...

@omarsar0: Big launch from @KeenableAI. The team built its own crawler, index, and query language for agents in...
@IPONewsroom_: KEENABLE RAISES $26M TO BUILD SEARCH INFRASTRUCTURE FOR AI AGENTS

Keenable has emerged from stealth...

7. Hesamation (Group Score: 326.4 | Individual: 56.1)

Cluster: 13 tweets | Engagement: 8758 (Avg: 411) | Type: Tech

> potential revenue: $30T > world GDP: ~$ 120T https://t.co/7S7v22iUb5\n\nQT @TheStalwart: According to WSJ, Anthropic will likely tell investors that its potential revenue is above $30 trillion. https://t.co/K76qzyvBml

See 12 related tweets

@TheStalwart: According to WSJ, Anthropic will likely tell investors that its potential revenue is above $30 trill...
@StockMKTNewz: ANTHROPIC IS EXPECTED TO TELL INVESTORS ITS POTENTIAL REVENUE OPPORTUNITY EXCEEDS $30 TRILLION

Ther...

@testingcatalog: ANTHROPIC 🔥: WSJ reports that Anthropic is pitching more than $30T in potential revenue to its inves...
@WhaleInsider: JUST IN: Anthropic is expected to tell investors it sees over $30 trillion in potential revenue - WS...
@cgtwts: investors after hearing Anthropic’s CEO say their potential revenue is $ 30T+ when the entire world’...

8. shiri_shh (Group Score: 319.4 | Individual: 43.0)

Cluster: 12 tweets | Engagement: 233 (Avg: 70) | Type: Tech

Anthropic market pitch: $30 TRILLION

Total US GDP: $32 TRILLION

The pre-IPO hype is now officially competing with national GDP numbers 😭 https://t.co/073KiYhWuj\n\nQT @kimmonismus: wtf Anthropic’s IPO pitch: more than $30 trillion in potential revenue, topping SpaceX and xAI’s$ 28.5 trillion estimate.

Anthropic may tell investors that the work Claude and other AI models could eventually perform represents a market worth more than $30 trillion.

The company could seek a valuation of roughly $2 trillion and raise up to$ 100 billion in its IPO.

I'm curious to see if that's possible after the current backlash. In any case, they'll need a good release before the IPO.

See 11 related tweets

@kimmonismus: wtf Anthropic’s IPO pitch: more than $30 trillion in potential revenue, topping SpaceX and xAI’s$ 28...
@HedgieMarkets: 🦔Anthropic is expected to tell IPO investors that its total addressable market exceeds $30 trillion....
@VaibhavSisinty: Anthropic the company behind Claude is about to go public. And they're telling investors their marke...
@DeItaone: ANTHROPIC EYES $30 TRILLION AI MARKET

Anthropic is expected to tell IPO investors its total address...

@IPONewsroom_: ANTHROPIC IS EXPECTED TO TELL INVESTORS ITS POTENTIAL REVENUE OPPORTUNITY EXCEEDS $30 TRILLION

The ...

9. huggingface (Group Score: 312.0 | Individual: 59.1)

Cluster: 8 tweets | Engagement: 724 (Avg: 88) | Type: Tech

RT @Apodex_AI: Meet Apodex 1.1: Scaling Agentic Intelligence for Complex Work

Open Source Harness: https://t.co/4V9suLx8o5 Open Weights: https://t.co/9CNZ9E5drH

We’re excited to introduce Apodex 1.1, our new model family built to scale agentic intelligence for professional work.

🧠 Frontier-level intelligence for complex work Apodex 1.1 brings frontier-level agentic performance across complex professional work, scientific research, financial analysis, and deep search.

🤝 Asynchronous Agent Team Apodex 1.1 can break down complex tasks, coordinate multiple agents in parallel, continuously integrate their findings, and let you step in to guide or redirect the work at any time.

🔬 Open-source research workbench We’re open-sourcing FrontierAgent—a locally deployable research workbench for the Apodex 1.1 family, including asynchronous Agent Team.

Available now: 🔹 Apodex 1.1 — our most capable frontier model, available through the Apodex online workbench 🔹 Apodex 1.1 mini — open-weight model for running complex work locally 🔹 FrontierAgent — open-source, locally deployable research workbench

🌐 Live workbench: https://t.co/w5l6YI2ZuP 🦾 API platform: https://t.co/kaxuSWvMB4 📃 Paper: https://t.co/CHc4L63U07

See 7 related tweets

@FellMentKE: The useful part of Apodex 1.1 is seeing complex work get decomposed, redirected, and reviewed instea...
@heyshrutimishra: 35B nearly matching 1T on agent benchmarks changes the cost math entirely.

Apodex 1.1 mini scores 2...

@Parul_Gautam7: Most AI tools give you an answer and leave the actual work to you.

What caught my attention about A...

@VaibhavSisinty: We have a new family of AI models now. And this one doesn't just think it works.

Apodex 1.1 launche...

@teortaxesTex: Apodex 1.1 mini is based on Qwen 3.5 35B A3B MoE btw\n\nQT @Apodex_AI: Meet Apodex 1.1: Scaling Agen...

10. MTSlive (Group Score: 303.9 | Individual: 37.2)

Cluster: 18 tweets | Engagement: 389 (Avg: 109) | Type: Tech

OpenAI the second they realized they could just make the chip themselves https://t.co/GfMt5xo1XZ\n\nQT @MTSlive: SITUATION DETECTED: OpenAI’s custom chip Jalapeño is showing significant performance gains over existing systems in internal testing. They plan to begin deploying it before the end of the year.

See 17 related tweets

@danielnewmanUV: $AVGO deserves a lot of credit here. And while I’ll reserve judgement on OpenAI silicon program for ...
@dylan522p: OpenAI Jalapeno is spicy Usually first generation chips aren't competitive, but OpenAI is beating Nv...
@MTSlive: SITUATION DETECTED: OpenAI’s custom chip Jalapeño is showing significant performance gains over exis...
@MTSlive: SITUATION UPDATE: OpenAI's models, including GPT-Astra, played a direct role in the development of J...
@haider1: openai is officially becoming a full-stack AI company

models: gpt-5.6 and beyond products: chatgpt ...

11. chris_j_paxton (Group Score: 262.0 | Individual: 32.3)

Cluster: 9 tweets | Engagement: 182 (Avg: 246) | Type: Tech

This is the mark of a general purpose robot foundation model -- that it can be taught tasks from examples without training.

We saw an early version of this from rhoda and a solid one last week from generalist, and now a really impressive, ten minute long one from skild. This to me does feel like the true gpt moment for robotics -- because if it truly generalizes, that means that you can really start deploying robots for anything, as developing robot skills becomes like prompting\n\nQT @SkildAI: Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning: https://t.co/wmF3Byv179

See 8 related tweets

@wallstengine: SKILD AI UNVEILS S1, A ROBOTICS MODEL THAT LEARNS FROM VIDEO

Its new S1 foundation model can learn ...

@VaibhavSisinty: What if you could teach a robot any task by showing it once? That's not a hypothetical anymore. 🤯

S...

@lightspeedvp: Could robots that keep learning after deployment be the path to general-purpose robotics?

@SkildAI ...

@heyshrutimishra: Robots used to need 100 hours of training data for one new task.

Skild AI just showed one that lear...

@StockSavvyShay: S1 just showed a really cool example of physical AI with a robotics foundation model that can learn ...

12. wallstengine (Group Score: 235.0 | Individual: 41.9)

Cluster: 9 tweets | Engagement: 542 (Avg: 200) | Type: Tech

More details on OpenAI’s Jalapeño chip from SemiAnalysis:

SemiAnalysis says Jalapeño is not narrowly optimized for OpenAI models.

It is a general-purpose inference ASIC, and in the workloads they tested it outperformed $NVDA,$ AMD and Google accelerators on several inference efficiency metrics.

The chip was designed from scratch for LLM inference and went from initial team buildout to tape-out in roughly 16 months.

Performance: • DeepSeek R1 exceeded 700 tokens/sec/user at concurrency 1

• Kimi K2.5 and GPT-OSS reached roughly 1,400 tokens/sec/user in selected tests

• Jalapeño achieved these results using single-token prediction, without speculative decoding or prefill/decode disaggregation

• SemiAnalysis says its output-token throughput per MW already exceeds the published July Vera Rubin results, despite Rubin using multi-token prediction

• On performance per total cost, Jalapeño and Vera Rubin are currently roughly comparable based on available results

These are still early benchmarks.

SemiAnalysis tested 8k/1k workloads and has not yet run its more demanding AgentX benchmark, which stresses long-context, multi-turn agentic workloads. Rubin is also already beginning to ship, while Jalapeño remains at the engineering-sample stage.

Hardware: • Built on TSMC N3P • 700W TDP • HBM4 with 15.4TB/s of memory bandwidth

• SemiAnalysis says that is currently the highest memory bandwidth per package among shipping/near-shipping accelerators

• A0 silicon produced the current benchmark results

• B0 is already in the fab and is expected to improve perf/W by roughly 25%

• B0 delivers 13.4 PFLOPs of MXFP4 versus 17.5 PFLOPs of dense NVFP4 for a similarly sized Rubin compute die, but at lower power

Scale: • 128 Jalapeño ASICs per rack • Up to 2,048 XPUs can be connected across a 16-rack scale-up domain

• A complete host + ASIC rack pair consumes roughly 160kW

• OpenAI’s next deployment target is around 100MW

OpenAI is also using Codex heavily in the chip’s development. SemiAnalysis says AI-assisted design reduced SIMD area by 8% and matrix-engine area by 10%, while Codex has been used to write and optimize kernels for new models.

Production is expected to ramp gradually through 2027, with most output currently planned for Q4 2027.\n\nQT @wallstengine: OpenAI says its new Broadcom-built Jalapeno AI chip outperformed $NVDA GB300 in both throughput per watt and response latency in its testing.

The chip is designed specifically for inference, not model training, and runs at around 700 watts. OpenAI plans to begin deploying it for its models later this year.

The company says the performance advantage widened on larger workloads, including Moonshot’s Kimi model, and that Jalapeno has also performed well internally on unreleased OpenAI models.

Important caveat: Jalapeno was tested against GB300, not NVIDIA’s newer Vera Rubin generation.

OpenAI says a second-generation chip is already approaching tape-out, while work on a third generation has begun.

Source: Bloomberg

See 8 related tweets

@zephyr_z9: WTF!!!\n\nQT @wallstengine: More details on OpenAI’s Jalapeño chip from SemiAnalysis:

SemiAnalysis ...

@StockSavvyShay: That OpenAI’s Jalapeño chip performance is coming from a 700W chip built with $AVGO on$ TSM N3P that...
@MilkRoadAI: RT @MelvinInvests: OpenAI just proved its homemade chip can outrun Nvidia's Blackwell and that's a g...
@cryptopunk7213: early results are out for openai’s custom ai chip, they went from design to production in just 9 mon...
@haider1: woah, openai's Gen 1 chip is already challenging Nvidia's Vera Rubin

on SemiAnalysis' InferenceX be...

13. _TheTathya (Group Score: 230.0 | Individual: 39.8)

Cluster: 12 tweets | Engagement: 257 (Avg: 120) | Type: Tech

Wow! This is HUGE 🔥

India Approves Transfer of All DRDO-Developed Conventional Missile Technologies to Domestic Industry 🇮🇳⚡️

India has approved the transfer of technologies for all DRDO-developed conventional missile systems to qualified Indian defence companies, paving the way for wider domestic production.

Missile technologies can now transition from DRDO development to industrial-scale manufacturing Indian private companies, MSMEs and technology partners will be able to participate Production will follow required qualifications, certifications and regulatory clearances The move will strengthen India’s domestic defence manufacturing capacity It is expected to increase indigenous value addition across the missile supply chain The policy will also help reduce dependence on imported defence systems and components

👉 India is moving from developing advanced missiles at home to building the industrial capacity to manufacture them at scale.

📌 Follow @_TheTathya for more such updates.

See 11 related tweets

@PIB_India: 🔰Raksha Mantri Rajnath Singh (@rajnathsingh) has approved the transfer of DRDO-developed technologie...
@TheMilObserverr: RT @TheMilObserverr: MoD has approved transfer of DRDO's technology for all conventional missile sys...
@CNBCTV18Live: #JustIn | Defence Minister #RajnathSingh approves the transfer of #DRDO-developed technologies of al...
@PIB_India: RT @DefenceMinIndia: In a significant step towards achieving the vision of Aatmanirbharta in defence...
@CNBCTV18News: Defence Stocks In Focus On DRDO Missile Tech Transfer
Defence Ministry approves transfer of DRDO-...

14. Dimillian (Group Score: 215.8 | Individual: 44.8)

Cluster: 7 tweets | Engagement: 764 (Avg: 111) | Type: Tech

RT @MatthewBerman: My full interview with Tibo (@thsottiaux)

0:45 Tibo's Lessons from Google DeepMind 4:22 Building OpenAI’s Relentless Culture 7:23 Astra & Next Gen Models 11:18 How Fast AI Changes Developer Workflows 14:27 ChatGPT & Codex Merging 20:25 OpenAI vs. Anthropic 23:37 Why OpenAI Keeps Resetting Limits 30:25 Recursive Self-Improvement 32:00 Dangers That Caused "The Pause" 34:13 Will Ultra Fast Become the Default? 43:20 Why Everyone Needs to Try AI

See 6 related tweets

@MatthewBerman: Tibo (@thsottiaux) talks about @OpenAI's culture:

"OpenAI has a very bottoms-up culture. It's a ver...

@MatthewBerman: Tibo on RSI:

"What we're seeing a ton of success with is using those models to develop the infrastr...

@MatthewBerman: Google had LMChat one year before ChatGPT launched.

Tibo (@thsottiaux): "I was part of that team....

@Gorden_Sun: 掌管Codex重置按钮的Tibo访谈，23:37讲的是重置的来历，基本上是否重置是Tibo一个人说了算，没有层层审批。这个岗位也太爽了。\n\nQT @MatthewBerman: My full i...
@derrickcchoi: Worth a watch! He does a lot more than just press buttons\n\nQT @MatthewBerman: My full interview wi...

15. OliviaReedai (Group Score: 210.5 | Individual: 31.3)

Cluster: 12 tweets | Engagement: 59 (Avg: 35) | Type: Tech

Wan 3.0 on Magnific looks seriously impressive.

30s generations with audio, up to 20 references, strong consistency and 1080p output.

This is getting very close to a complete production workflow inside one tool.\n\nQT @magnific: From individual shots to full scenes, here's Wan 3.0

→ 30s generations, audio included → Up to 20 references → Pixel-level consistency → 1080p, five ratios or adaptive framing

Standard, or Prime when you need it faster Up to 40% off

Now on Magnific https://t.co/dPQMNVz7lL

See 11 related tweets

@NovaIAHQ: The reference consistency is what caught my attention here.

Up to 20 references in a single generat...

@TheoBuildsAI: The quality of AI video models is improving, but control is becoming just as important.

Wan 3.0 on ...

@SynapseOpsAI: One of the biggest problems with AI video has always been keeping characters and scenes consistent. ...
@TheoBuildsAI: RT @OliviaReedai: Wan 3.0 on Magnific looks seriously impressive.

30s generations with audio, up to...

@ethancole_ai: Magnific adding Wan 3.0 makes a lot of sense.

One platform for generating, refining and improving t...

16. omarsar0 (Group Score: 205.1 | Individual: 35.9)

Cluster: 9 tweets | Engagement: 35 (Avg: 56) | Type: Tech

Can't wait to have always-on agents running locally 24x7. Local agents are something Perplexity seems to be chasing these days aggressively, and it makes a ton of sense. There is a huge market for personalized AI agents.\n\nQT @nvidia: Meet Portable Computer, Perplexity's new local-first agent stack on NVIDIA DGX Spark.

When running locally, Portable Computer offers one-click local inference setup and an optimized agentic experience for DGX Spark.

Learn more and get started today: https://t.co/0UElxhCnkY https://t.co/W07ZafKYdC

See 8 related tweets

@NVIDIARTXSpark: Congrats to @Perplexity_AI on launching Portable Computer optimized to run locally on DGX Spark! 🎉

...

@NVIDIAAI: RT @nvidia: Meet Portable Computer, Perplexity's new local-first agent stack on NVIDIA DGX Spark.

W...

@PaulSolt: RT @nvidianewsroom: The first CPU built for agents is now at work at massive scale.

@SpaceXAI is de...

@AravSrinivas: Big thanks to @nvidia for working together with us and supporting this research on DGX Spark as well...
@MSBIntel: JUST IN: Perplexity and Nvidia launch Portable Computer, running AI locally on Nvidia hardware....

17. josevalim (Group Score: 175.8 | Individual: 36.5)

Cluster: 9 tweets | Engagement: 814 (Avg: 100) | Type: Tech

RT @tobi: I’m thinking about banning Claude code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.

Insisting on only reading CLAUDE.md sometimes leads to split brain problems when different team members use different tools. Just unnecessary.

See 8 related tweets

@dani_avila7: Not following the AGENTS.md standard and sticking with CLAUDE.md is becoming a problem at scale

I h...

@julien_c: Hoping tobi’s tweet has enough visibility to make them change this!\n\nQT @tobi: I’m thinking about ...
@omarsar0: Glad Tobi posted about this.

This issue has resurfaced countless times, and they have done nothing ...

@TMTLongShort: Everyone has principles until Ant drops a new model that mogs SOTA and all of a sudden ZDR stands fo...
@Daniel_Farinax: How about CI to ban Claude.md from all repos and a system wide enterprise rule to force read Agents....

18. wallstengine (Group Score: 173.0 | Individual: 28.2)

Cluster: 10 tweets | Engagement: 685 (Avg: 200) | Type: Tech

$GOOGL LAUNCHES GEMINI ENTERPRISE FOR LEGAL

Google Cloud is launching Gemini Enterprise for Legal in preview, with Cleary, Freshfields, Weil and Williams & Connolly among initial customers.

The platform targets contract review, regulatory monitoring, legal brief drafting, citation verification and data requests, with integrations including DocuSign, Thomson Reuters, Harvey, iManage and RelativityOne.

Google says customer data, prompts and outputs remain within each organization’s private cloud environment and are not used to train its base models.

See 9 related tweets

@testingcatalog: Google announced its first plugins for Gemini Enterprise, tailored for legal and finance use cases. ...
@StockMKTNewz: Google Cloud $GOOGL just unveiled a bunch of new partnerships for its new AI features targeted at Le...
@StockSavvyShay: $GOOGL is taking Gemini deeper into enterprise software with a new legal platform for contract revie...
@Reuters: Google expands Gemini Enterprise AI platform for law firms, lawyers https://t.co/Jwm8WydV4d https://...
@RoundtableSpace: Google just launched Gemini Enterprise for legal and financial services

There is truly no safe ind...

19. KatieMiller (Group Score: 169.1 | Individual: 41.8)

Cluster: 5 tweets | Engagement: 10550 (Avg: 3379) | Type: Tech

Cloudseeding has left the conspiracy theory chat\n\nQT @ADoricko: Rainmaker just produced ~19M gallons of water in Alaska via next-gen cloud seeding over 3 hours of operations.

We are the first company to provably produce precipitation in Alaska. As promised, we’ve linked our white paper and relevant data.

In the future, Rainmaker will protect and restore glaciers with man-made snowfall.

Immediately, this demonstration shows how Rainmaker will add new water to the Colorado River and Great Salt Lake in the coming months.

We turned around this analysis and white paper just a couple days after operating. There are many data sources I want to deepen our understanding of the atmosphere and increase our provable production; we’re building the instruments to collect that data in subsequent operations. But at Rainmaker, we won’t tout LOIs, simulations, or lab tests. What matters is physically measuring that you’ve affected the atmosphere; proving that we’ve produced water. We prefer sharing only the tech that has proven to make more water for farms, industries, and ecosystems in need.

Expect a regular cadence of these from us as we build the best atmospheric science lab in the world. Feedback is welcome in the interim.

Rainmaker Century.

https://t.co/QZikxP1KyH

See 4 related tweets

@TaylorLorenz: Rainmaker is one of the coolest technology companies out there. Truly sci fi level accomplishment\n...
@APompliano: Rainmaker is proving it can create man-made snowfall, which could protect the glaciers in Alaska.

...

@LongJourneyVC: Rainmaker is the first company to provably produce precipitationin Alaska.

We repeat: Rainmaker is...

@paulg: RT @ADoricko: Rainmaker just produced ~19M gallons of water in Alaska via next-gen cloud seeding ove...

20. BrianRoemmele (Group Score: 166.9 | Individual: 36.8)

Cluster: 6 tweets | Engagement: 117 (Avg: 257) | Type: Tech

EVEN OPENAI GAVE UP WITH “AI DETECTOR” GRIFT

OpenAI itself tried building a classifier and then shut it down because the accuracy was garbage.

They admitted it labeled human text as AI 9 percent or much more of the time while only catching a fraction of actual AI text.

The company that built the models most people worry about could not make a detector that worked. That should have been the end of the conversation.

Instead a whole industry of grifting companies seeking to institutions like PENTAGRAM, GPTZero, Originality . ai, ZeroGPT, and the rest just kept selling the fantasy.

The companies and their “Slop Janitor”!shills are peak arrogance. They sell fear. They sell the idea that without their product the academy or the workplace will collapse into AI sludge.

Then many of them turn around and secretly offer “humanizer” tools that take AI text and rewrite it until the same detectors they sell say it is human.

Get it?

The cycle is perfect grift. Create the panic. Sell the detector. Sell the way around the detector. Repeat.

Read more below:\n\nQT @BrianRoemmele: A MODERN DAY AI WITCH HUNT

Today good thoughtful writers are being accused of being guilty of "AI slop" writing and this is causing great writers to dumb down their work, all for the grifters selling “detector tools” as solutions.

https://t.co/43WWykAT8X

See 5 related tweets

@BrianRoemmele: HOW THE “AI DETECTOR” GRIFT WORKS—THE GATEKEEPERS OF KNOWLEDGE

Here is the real game. These detecto...

@BrianRoemmele: The AI Detector Grift

Why the algorithms policing human writing are built on bad science, arrogance...

@BrianRoemmele: RT @BrianRoemmele: EVEN OPENAI GAVE UP WITH “AI DETECTOR” GRIFT

OpenAI itself tried building a clas...

@BrianRoemmele: BOOM! THANKS A MILLION!

This article on a clueless company and their worthless “AI Wrote This Detec...

@rohanpaul_ai: RT @BrianRoemmele: A MODERN DAY AI WITCH HUNT

Today good thoughtful writers are being accused of b...