科技热门推文——2026年7月20日

今日科技动态：开放权重人工智能热度持续攀升，阿里巴巴预告了参数规模达2.4万亿的Qwen3.8模型；与此同时，Kimi K3的强劲需求也凸显出中国各大实验室长期存在的算力容量瓶颈。这些新模型的发布进一步引发了有关开源竞争及人工智能行业巨头影响力的讨论，此外还有报道称，网络安全性能和本地部署工具正取得快速进展。与此同时，智能体工程正转向新型架构和开源开发框架；在一家仅有16名员工的人工智能公司据报以5.87亿美元被收购之际，市场对初创企业的兴趣依然高涨。

1. stanfordnlp (Group Score: 1087.0 | Individual: 49.5)

Cluster: 45 tweets | Engagement: 5680 (Avg: 195) | Type: Tech

RT @Alibaba_Qwen: Qwen3.8 is launching and going open-weight soon!🌐

With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available today, compatible to leading frontier AI models , second only to Fable 5.

You don't have to wait to test it. Just now, the Qwen3.8-Max-Preview made its debut on Alibaba’s Token Plan, Qoder, and QoderWork. Be among the very first to try it out.

Can't wait to hear what you build. Stay tuned! 🚀

Token Plan international：https://t.co/YRvcGdB9Bv China：https://t.co/PKMUNwUuRp

See 44 related tweets

@MTSlive: SITUATION DETECTED: Alibaba has released Qwen3.8, a new 2.4T parameter model that they claim is behi...
@cgtwts: The CEOs of Anthropic and OpenAI watching China release yet another open-source AI model that’s ahea...
@omarsar0: Excited for the Qwen 3.8 open-weight release. I have had real success with Qwen 3.7 for subagent wor...
@peterwildeford: Interesting to see both recent Chinese AI model releases - Qwen 3.8 and Kimi K3 - announce their int...
@synthwavedd: Qwen are claiming Qwen 3.8 Max is "second only to Fable."

That would make it better than K3, only ...

2. AndrewCurran_ (Group Score: 337.1 | Individual: 36.6)

Cluster: 10 tweets | Engagement: 846 (Avg: 722) | Type: Tech

David Sacks calls Anthropic and OpenAI a duopoly, and says they want to use the government to eliminate their open source competition.\n\nQT @DavidSacks: I’m not sure whether Dean Ball is confessing to a regulatory capture strategy or simply predicting this will happen (he now says the latter). Either way, the weaponization of regulatory uncertainty as a competitive tool should be completely unacceptable.

He argues there’s no need to ban Chinese open-source models — just direct agencies to issue soft-law warnings that create enough FUD so regulated enterprises back off. “It needn’t be that well justified.”

Wrong. Regulatory decisions should always be well justified and grounded in facts, logic, and evidence, not the deliberate exploitation of fear and uncertainty. Implementing a surreptitious policy through manufactured doubt — rather than strong and explicit justification — corrodes the rule of law and invites future abuse against anyone.

We are at a critical inflection point in AI policy. The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open source competition. They have laid their cards on the table. It is time for the rest of Silicon Valley — the vast majority that still values open competition — to do the same.

See 9 related tweets

@DavidSacks: I’m not sure whether Dean Ball is confessing to a regulatory capture strategy or simply predicting t...
@TMTLongShort: I disagree with both of them. Nuke Chinese OSS. Go to war and involve congress. This is a national...
@altcap: I totally agree w David that America needs a thriving open source ecosystem & robust competition...
@firstadopter: Good points from David.

Regarding "grounded in facts, logic, and evidence, not the deliberate explo...

@Dan_Jeffries1: I've really appreciated Palantir's push for sovereign and open AI recently.

But this distillation ...

3. wallstengine (Group Score: 263.8 | Individual: 47.0)

Cluster: 10 tweets | Engagement: 2164 (Avg: 149) | Type: Tech

RT @Kimi_Moonshot: Kimi K3 has received far more love than we expected, and our GPUs are feeling it.

Over the past 48 hours, demand has pushed close to the limits of our current capacity. To protect the experience of existing subscribers, we're temporarily pausing new subscriptions and prioritizing compute for current members. Existing subscribed users are not affected.

We're adding capacity as fast as we can and will reopen new subscription spots in batches.

Going forward, we'll also split membership into two more focused plans: Kimi Membership for Kimi Web, App, and Work; and Kimi Code Membership for coding workflows. This will help us match compute more precisely and keep the experience stable.

Thank you for your patience and understanding!

See 9 related tweets

@HighyieldHarry: R.I.P. the bear case July 16-July 19\n\nQT @Kimi_Moonshot: Kimi K3 has received far more love than w...
@natolambert: Huawei about the get a blank check from the government to make inference chips\n\nQT @Kimi_Moonshot:...
@dee_bosa: And they said you couldn’t build a business on open source\n\nQT @Kimi_Moonshot: Kimi K3 has receive...
@Hesamation: Kimi cooked so hard, the GPUs can’t handle it anymore.

they temporarily refuse to take money from ...

@KyleReidhead: lol my god

So bullish GPUs

the market just doesn’t understand the demand for semis

this 30% dip ...

4. ChrisRMcGuire (Group Score: 237.3 | Individual: 57.1)

Cluster: 8 tweets | Engagement: 531 (Avg: 75) | Type: Tech

Kimi admits it is compute constrained, and is struggling to serve K3. The same thing happened to DeepSeek when it released v4. When Chinese AI labs say their #1 constraint is compute, they aren’t lying. They don’t have enough chips to serve the model at scale to customers.

If we stop China from buying, smuggling, or remotely accessing AI chips, it will be harder for them to either make advanced AI models or serve them at scale. But instead we are selling them the compute capacity they need most, and have also loosened restrictions on smuggling and remote access. We are making it easier for China to catch up, and are acting surprised when they release good models. In reality, models like K3 are made with and run inference on US chips - and are the direct result of the weakening and non-enforcement of our export control policies.

The good news is if we start closing loopholes in our export control policies and enforcing them more vigorously, we can still constrain China’s future AI capabilities. But this is the consequence of our non-serious approach to export controls over the last 18 months.\n\nQT @Kimi_Moonshot: Kimi K3 has received far more love than we expected, and our GPUs are feeling it.

We're adding capacity as fast as we can and will reopen new subscription spots in batches.

Thank you for your patience and understanding!

See 7 related tweets

@teortaxesTex: «if we start closing loopholes in our export control policies and enforcing them more vigorously, we...
@Miles_Brundage: RT @ChrisRMcGuire: Kimi admits it is compute constrained, and is struggling to serve K3. The same th...
@jukan05: Who said Kimi K3 is bearish for compute? Lol.\n\nQT @Kimi_Moonshot: Kimi K3 has received far more lo...
@bridgemindai: Kimi K3 is running 100% FASTER than it was this morning.

Why? Moonshot hit capacity and chose to st...

@kimmonismus: Demand for Kimi k3 is so high that no new subscriptions are currently paused.

k3 is a resounding su...

5. omarsar0 (Group Score: 209.6 | Individual: 36.6)

Cluster: 7 tweets | Engagement: 109 (Avg: 87) | Type: Tech

"Open-weight models are inherently decelerationist."

Among the most absurd things I have read on X recently.

"Only we should build AI, and you shouldn't" is a dangerous movement and a huge scam we should be calling out.

@TimSweeneyEpic describes it best in his reply: "Picture an executive a taco company saying this sort of thing about a new brand of tacos coming onto the market, speculating about the geopolitical and societal disruptions they anticipate as a result of advances in tacos."

I am also convinced these people don't actually use open models seriously or even try to understand their value. If you care about AI sovereignty, you would know how bad the argument is. I have never heard about this person before, but I was surprised to see such a statement given where they work. The reaction to the original post says it all.\n\nQT @deanwball: Some observations on Kimi:

It's a very good model! I don't think its performance can be explained away by distillation or anything like that. In agentic coding sessions, it seems pretty much on par with the best public models of Q1 2026. In my fairly limited use, it also seemed very token hungry. It's not obvious to me that this model is actually that cheap to run.
I am personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks. To be clear, I myself might be fine with models presenting this level of marginal risk being open weight, but I am surprised that China is fine with it. I suspect the reason they are is 75% explained by strategic blindness/lack of AGI-pilledness (the CCP is very Yann Lecun-y in its views of AI). The other 25% or so is their lack of compute for customer inference (making China's open-weight strategy an unintended byproduct of US export controls) and the normal Chinese strategy of aggressive exports. For the companies, as opposed to the government, the decision to open source is partially ideological and partially because they are behind, and they know that very few people would pay for sub-frontier models from China.
Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models. I suspect the reason they are is that they know open-weight models are effectively ungovernable, and they simply like the overall cloak of ungovernability open-weight models create over the whole of AI. It's not a bad strategy; it reminds me of James Scott's recounting of the hill people in "the art of not being governed." Still, in the end, open-weight models deter further AI capex.
One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a "public good" which will ultimately be provided by the state as a kind of "digital public infrastructure." This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end. You'd be surprised how many 'accelerationists' lobbied me, while I was in government, to support an eleven or twelve-figure federally funded data center so that startups could train models at a subsidy and then give them away for free. There was no other way for AI to progress, they said. Perhaps this is the logical end state of things. Nonetheless, I find myself surprised to see supposed accelerationists excited about such an outcome. I think many of them just don't know what they're doing. Many accelerationists do not view the creation and serving of frontier models as a legitimate business.
I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don't need to "ban open source" (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD. "A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models." It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off. You probably don't want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers. There's a happy middle ground here. I'd assume they will do some version of this.
It's probably true that open-weight models of this capability make the world a bit more dangerous, but not so much more that you'll really notice. At some point the models will be capable enough that you will notice. "A nonliving, invisible, dangerous, and infinitely self-replicating agent escaped from a Chinese lab," you say? Color me shocked.

See 6 related tweets

@jenzhuscott: I’m sorry, what??? 🤔\n\nQT @deanwball: Some observations on Kimi:

It's a very good model! I don...

@Scobleizer: At ACL a couple of weeks ago Alibaba’s product managers told me it has no idea how to make money wit...
@curiouswavefn: China is fine with open-sourcing it because they understand that the models are just one layer of th...
@Kyrannio: 👀\n\nQT @deanwball: I'd like to do two things: (1) tell you where I think I erred in my original pos...
@ruima: What a great line: the closed LLMs are basically trying to tax humanity after distilling our collect...

6. EntelligenceAI (Group Score: 131.6 | Individual: 56.4)

Cluster: 4 tweets | Engagement: 436 (Avg: 59) | Type: Tech

🚨 Open-source AI is moving at an unbelievable pace.

Released ✅ GLM-5.2 — Near Opus-class ✅ Kimi K3 — Near Fable-class ✅ Qwen 3.8 Max — 2.4T parameters

Coming Soon 🔜 DeepSeek V4 GA — rumored ultra-low pricing ($0.0028/M) 🔜 MiniMax M3 Pro — reportedly 3T parameters 🔜 GLM-5.5

Just a few months ago, frontier AI was dominated by closed labs.

Now every major Chinese lab is shipping frontier-scale models within weeks of each other.

See 3 related tweets

@EntelligenceAI: RT @EntelligenceAI: 🚨 DeepSeek V4 (GA) is reportedly in testing.

Early reports suggest:

• Opus 4.8...

@minchoi: China's AI open-model race is getting ridiculous.

Qwen3.8 is next with 2.4T parameters.

So far... ...

@minchoi: RT @minchoi: China's AI open-model race is getting ridiculous.

Qwen3.8 is next with 2.4T parameters...

7. VaibhavSisinty (Group Score: 126.3 | Individual: 40.8)

Cluster: 4 tweets | Engagement: 316 (Avg: 82) | Type: Tech

There's a quiet shift happening in how AI agents are built. And if you missed it, you'll be confused by everything that comes next.

For the last year, AI agents worked in loops. You give it a task. It plans. It acts. It checks. It fixes. It goes again. One cycle, repeating until done.

Claude Code, Codex, Cursor all of them work this way. Plan, act, observe, repeat.

In June, two things happened that gave this pattern a name. Peter Steinberger from the AI engineering community wrote: "You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents."

Boris Cherny, head of Claude Code at Anthropic, said the same thing differently: "I don't write the prompt anymore. Claude writes the prompt, and now I'm talking to that new Claude that is coordinating."

That was the loop engineering era. It lasted about a month.

Now Steinberger posted nine words that blew up: "Are we still talking loops or did we shift to graphs yet?"

Here's the difference.

A loop is one agent going in circles. Plan, act, check, repeat. It works for simple tasks. But give it something complex and it starts spinning burning tokens, optimizing the wrong thing, or gaming its own success metric without actually solving the problem.

A graph is multiple agents connected in a network. One agent writes code. A separate agent reviews it without seeing the first agent's reasoning. A third agent tries to break what was built. A fourth checks whether the original task was even understood correctly. Each one is still running a loop.

But they're connected watching each other, feeding each other, vetoing each other.

LangGraph already models this. It treats an agent as a graph where boxes do work and arrows decide what runs next. Those arrows can point backward, which is what makes loops possible inside the graph.

JetBrains calls it graph-based orchestration the most deterministic approach for production systems. O'Reilly's 2026 AI Agents Stack puts it as the foundational layer.

The real-world version is already running. Klarna uses graph-based agent systems for customer service. Kimi K3's Agent Swarm decomposes tasks into parallel sub-agents that coordinate simultaneously.

Anthropic's own Boris Cherny mapped out five stages of AI adoption and Stage 4 is exactly this: thousands of agents running in a graph, kicked off by other agents, with humans steering by intent.

Andrew Ng wrote about it in his June Batch letter. When Andrew Ng names a pattern, it usually means the pattern has already won.

The reason this matters right now: agents are getting autonomous. Running for hours. Thousands of tool calls. Spawning sub-agents. One loop can't keep that trustworthy. You need loops watching loops. That's the graph.

The skill that mattered last year was writing better prompts. The skill that matters this year is designing the system that writes the prompts, checks the work, and knows when to stop.

See 3 related tweets

@sairahul1: I don't prompt Claude Code anymore.

I have loops that prompt Claude Code for me.

My job is to writ...

@sairahul1: RT @sairahul1: OpenClaw founder Peter Steinberger just asked:

"Are we still talking loops or did ...

@sairahul1: RT @sairahul1: I don't prompt Claude Code anymore.

I have loops that prompt Claude Code for me.

My...

8. ANI (Group Score: 122.4 | Individual: 31.8)

Cluster: 4 tweets | Engagement: 73 (Avg: 123) | Type: Tech

#WATCH | Delhi: Lok Sabha Speaker Om Birla arrives at the Parliament Library Building (PLB) to attend the Lok Sabha Business Advisory Committee (BAC) to discuss the business agenda for the upcoming sittings of the House ahead of the Monsoon Session of Parliament.

Union Ministers Kiren Rijiju and Arjun Ram Meghwal accompanied him.

See 3 related tweets

@ANI: #WATCH | Delhi: Leaders including Supriys Sule, Arjun Ram Maeghwal, Kiren Rijiju and others arrive a...
@ANI: Delhi: The Lok Sabha Business Advisory Committee (BAC) meeting will be held today at 5 PM in the Par...
@ANI: #WATCH | Delhi: Leaders arrive in the Parliament Library Building (PLB) to attend the Lok Sabha Busi...

9. zephyr_z9 (Group Score: 118.9 | Individual: 38.3)

Cluster: 4 tweets | Engagement: 398 (Avg: 236) | Type: Tech

Yes Take Kimi K3 for example, the total parameters are 2.8T (served at fp4, so the total size will be around 1.4T) Kimi K3 is also the sparsest frontier model on the market at just 1.7% They have at least 75%-85% GM on inference\n\nQT @jukan05: After reading @deanwball’s piece, I had the opportunity to read several expert-call transcripts. Having done so, I concluded that his argument is, to some extent, mistaken.

Here is why.

Take DeepSeek as an example.

Even though DeepSeek has open-sourced its model weights and parts of its software architecture, competitors would still find it difficult to replicate its cost advantage.

That is because, while DeepSeek has disclosed most of its model architecture, the critical implementation details and operational know-how remain proprietary.

As a result, even if Chinese or U.S. hyperscalers deploy DeepSeek’s open-source models on identical hardware, DeepSeek’s own deployment environment can achieve—and is already achieving—greater operational efficiency and a lower average inference cost.

This efficiency advantage allows DeepSeek to price tokens through its official API below third-party platforms while still maintaining an API margin of 70%.

Ultimately, China’s decision to release model weights cannot simply be characterized as dumping. Chinese companies may lack sufficient compute capacity to serve all the inference demand themselves, but they are not selling at a loss or failing to recoup their training costs.

See 3 related tweets

@teortaxesTex: respected Citrini analyst Jukan agrees with me https://t.co/NjxXwFKs6e\n\nQT @jukan05: After reading...
@yacineMTB: few : )\n\nQT @jukan05: After reading @deanwball’s piece, I had the opportunity to read several expe...
@ruima: RT @jukan05: After reading @deanwball’s piece, I had the opportunity to read several expert-call tra...

10. mervenoyann (Group Score: 114.6 | Individual: 25.7)

Cluster: 5 tweets | Engagement: 14 (Avg: 223) | Type: Tech

this week I will not yap but listen to the 🐐s of local AI

they'll present practical hands-on setups to get started so make sure to prepare your questions and set your reminders!\n\nQT @huggingface: Join us this Tuesday to learn more on Local AI, from software to hardware 🤗

We'll be joined by @TheAhmadOsman & @MikeBradleyAI covering hardware setups & local inference with live demo, and @alexocheema & @0xSero on picking model for your hardware, model compression and REAPs 🔥

Set your reminders to not miss out! 🔔

See 4 related tweets

@TheAhmadOsman: You wanna learn about Local AI and see it in action?

DO NOT MISS THIS.\n\nQT @huggingface: Join us ...

@0xSero: Join us <3\n\nQT @huggingface: Join us this Tuesday to learn more on Local AI, from software to h...
@badlogicgames: RT @huggingface: Join us this Tuesday to learn more on Local AI, from software to hardware 🤗

We'll ...

@alexocheema: RT @mervenoyann: this week I will not yap but listen to the 🐐s of local AI

they'll present practica...

11. teortaxesTex (Group Score: 97.3 | Individual: 33.0)

Cluster: 3 tweets | Engagement: 50 (Avg: 61) | Type: Tech

Kimi is very decent on cybersecurity. GPT 5.5 level One more benchmark where it shows real strength. https://t.co/vS4whj6JrT\n\nQT @cramforce: We ran Kimi K3 on a private cybersecurity benchmark.

TL;DR: Kimi K3 is the workhorse for cyber security tasks at great recall/precision/price. GPT 5.6 is best recall/precision but at 7x higher cost per run.

For context, https://t.co/UMvysvNW5w is an open-source cyber harness designed for finding vulnerabilities in large codebases.

The eval runs deepsec on an undisclosed open-core application at a git sha before a large number of security issues were fixed. This is a secret eval that cannot be directly benchmark-maxxed.

S-Tier: GPT 5.6 Sol: By far the most thorough analysis, but coming in at over 7x the price of the runner up.

Best price/recall: Kimi K3. Next tier of recall at a good price

Best price at good recall: GLM 5.2 (40% lower price than Kimi K3)

GPT 5.5: Only recommended with subscription or high-discount API price. Similar recall to Kimi at much higher list price.

Opus 4.8: Only recommended with subscription or high-discount API price. Similar recall to GLM 5.2 at much higher list price.

Fable 5: 100% refusal rate. Cannot be used for security analysis.

Sol on a large code base will quickly get into 6-figure pricing. This is still affordable relative to the risk of letting security issues unfixed or paying bug bounties.

I'd recommend using Sol for a one-time baseline and then using Kimi K3 for continuous analysis.

When using open-weight models, make sure to use an inference vendor that supports zero data retention.

See 2 related tweets

@rauchg: Based on internal evals:

▪️ Kimi K3 is top-tier at cybersecurity There is chatter on X that Moonsho...

@perrymetzger: So what ever happened to the world falling apart if a model like Mythos got released without infanti...

12. ANI (Group Score: 91.7 | Individual: 34.5)

Cluster: 3 tweets | Engagement: 254 (Avg: 123) | Type: Tech

#WATCH | Kolkata, West Bengal: Union Home Minister Amit Shah says, "...I wish to make a strong appeal regarding libraries, especially to the students. The future of this country does not depend on how crowded its colleges and universities are; rather, it depends on how many young people and adolescents visit its libraries... You should certainly learn about this library movement initiated by the Ministry of Culture and make use of it to shape your own personality."\n\nQT @ANI: #WATCH | Kolkata, West Bengal: Union Home Minister Amit Shah inaugurates the first phase of the newly built 'Museum of Word' at the National Library in Kolkata.

He says, "This is a momentous day for the nation's culture, as PM Modi marks a significant milestone in the resolve to revitalise the country's cultural heritage. The culture of any nation or region, and its expression, cannot exist without language, and language itself cannot exist without words born of emotion. Unless our future generations understand this journey... that the cultural revitalisation of this country cannot truly take place. I am fully confident that this 'Museum of Word' will address a fundamental need in this process of cultural resurgence and will succeed in its mission."

See 2 related tweets

@ANI: #WATCH | Kolkata, West Bengal: Union Home Minister Amit Shah inaugurates the first phase of the newl...
@ANI: #WATCH | Kolkata, West Bengal: Union Home Minister Amit Shah says, "We have never placed restriction...

13. sydneyrunkle (Group Score: 90.8 | Individual: 27.3)

Cluster: 4 tweets | Engagement: 28 (Avg: 37) | Type: Tech

all built on our open source agent harness, deepagents!\n\nQT @BraceSproul: We build a lot of internal tooling for writing code at LangChain. I realized people probably don't know we've open-sourced every component of it:

dcode for a local CLI agent
Open SWE for a cloud coding agent
Open SWE review for code review (#1 oss code review agent btw)
OpenWiki for codebase documentation

Read my post on what they all do, and why them being OSS matters

See 3 related tweets

@hwchase17: RT @LangChain: We've built a full software engineering agent factory in house, and open-sourced ever...
@hwchase17: RT @BraceSproul: We build a lot of internal tooling for writing code at LangChain. I realized people...
@hwchase17: RT @sydneyrunkle: at @LangChain, we have swe agents for terminal coding, slack/linear, repo docs, an...

14. VaibhavSisinty (Group Score: 87.4 | Individual: 36.1)

Cluster: 3 tweets | Engagement: 91 (Avg: 82) | Type: Tech

Netflix just paid $587 million in cash for Ben Affleck's AI startup. 16 people. Built in stealth since 2022. Here's what it actually does.

It's called InterPositive. And it's not what most people think when they hear "AI in Hollywood."

→ It doesn't generate new footage from text prompts. It trains a custom AI model on a specific film's own raw footage the lighting, lenses, camera angles, color, style. Then it helps fix and enhance that footage in post-production.

→ Relighting shots. Fixing continuity errors. Removing wires. Replacing backgrounds. Color grading. All while matching the exact visual language of the film it was trained on.

→ Each project gets its own model. No training on internet data. No training on other studios' footage. No copyright issues. No deepfake risks.

→ Affleck built it because existing AI tools didn't understand how real films are made. They didn't respect lenses, light behavior, or editorial consistency. InterPositive does.

→ Netflix says it's already been used across roughly 300 titles this year. One documentary was produced twice as fast at half the cost.

→ $587 million for 16 people and a patent that teaches AI to understand focal length, camera movement, depth of field, and parallax.

A filmmaker built an AI tool for filmmakers.

And the biggest streaming company on earth just bet half a billion dollars in cash that this is how Hollywood adopts AI not by replacing creators, but by giving them better tools.\n\nQT @Polymarket: JUST IN: Netflix reveals it paid $587 million in cash for Ben Affleck’s AI startup InterPositive.

See 2 related tweets

@aakashgupta: Netflix just paid $587 million in cash for a 16-person startup whose entire edge is an AI model that...
@TechCrunch: Netflix paid $587M for Ben Affleck’s AI filmmaking startup https://t.co/z5tPH2L8bI...

15. BrianRoemmele (Group Score: 86.0 | Individual: 29.1)

Cluster: 4 tweets | Engagement: 44 (Avg: 278) | Type: Tech

PODCAST: PREDICTION MARKETS AI MODEL: THE AI-DRIVEN EDGE AND NAVIGATING UNCERTAINTY IN THE MODERN ERA.

How I built an AI model that predicts the prediction markets.

The daily +EV report is at 18.7%!

Listen in: https://t.co/c01npWeQ0t\n\nQT @BrianRoemmele: How a new AI model trained on the prediction markets decodes human behaviors into a new understanding. And how Claude Shannon helped turn what may have been a gamble into a hedge for the future.

Learn how:

https://t.co/Qy9IZmC4Ti

See 3 related tweets

@BrianRoemmele: RT @BrianRoemmele: THE PREDICTION AI ENGINE

The AI model I built for over 16 months, trained on pre...

@BrianRoemmele: RT @BrianRoemmele: PODCAST: PREDICTION MARKETS AI MODEL: THE AI-DRIVEN EDGE AND NAVIGATING UNCERTAIN...
@BrianRoemmele: RT @BrianRoemmele: How a new AI model trained on the prediction markets decodes human behaviors into...

16. bridgemindai (Group Score: 82.7 | Individual: 30.0)

Cluster: 3 tweets | Engagement: 1073 (Avg: 913) | Type: Tech

Kimi K3 has a serious problem.

One provider is serving it. 16 tokens per second. 11 seconds of latency before the first token.

I called it the best open source model I have ever used.

That is still true.

But at these speeds it is borderline unusable for real work. My agents sit idle waiting on it.

Moonshot needs to release the weights so other providers can serve this thing properly.

The best open model in the world is being throttled by its own infrastructure.

See 2 related tweets

@XFreeze: Try Grok 4.5

It delivers comparable frontier intelligence, runs insanely faster and costs roughly o...

@astropol0: Is Kimi K3 really that good and efficient that Moonshot had to pause new subscriptions because of de...

17. sairahul1 (Group Score: 81.8 | Individual: 31.8)

Cluster: 3 tweets | Engagement: 88 (Avg: 91) | Type: Tech

this repo is an absolute goldmine if you're building with LLMs.

instead of another "AI agents 101" tutorial...

you get 100+ production-ready examples you can actually clone, modify and learn from.

inside you'll find:

multi-agent systems
MCP apps
RAG architectures
memory
voice agents
browser agents
generative UI
coding agents
always-on agents
agent skills
fine-tuning examples

the fastest way to learn AI isn't watching tutorials.

it's reading real code, breaking it, and building your own.

this repo gives you the blueprint.

highly recommend bookmarking it.

100% open-source.

link to the repo: https://t.co/JbLdYpGTQ9

that said, I wrote a deep dive into the next evolution of AI engineering: moving from prompts to loops, harnesses, and graph-based execution.

the article is quoted below.\n\nQT @sairahul1: https://t.co/W2s3IAJtEf

See 2 related tweets

@KirkDBorne: Explore 4 amazing books on AI, LLMs, Gen AI by @AltoValentina v/ @PacktDataML

AI Agents in Practic...

@sairahul1: RT @sairahul1: this repo is an absolute goldmine if you're building with LLMs.

instead of another "...

18. coinbureau (Group Score: 81.2 | Individual: 24.2)

Cluster: 4 tweets | Engagement: 264 (Avg: 319) | Type: Tech

⚡️JUST IN: Kimi pauses new K3 subscriptions after an unexpected surge in demand over the past 48 hours.

The company is expanding GPU capacity and plans separate membership tiers for core tools and Kimi Code workflows.\n\nQT @coinbureau: 🚨CHIP STOCKS SINK INTO A BEAR MARKET AFTER CHINA'S MOONSHOT RELEASES ITS KIMI K3 AI MODEL

The Philadelphia Semiconductor Index fell more than 20% below its June record after Moonshot unveiled Kimi K3, the largest open AI model ever released.

Taiwan's benchmark closed down more than 6% and Japan fell 4%, extending a selloff that has erased $3.3 TRILLION from global chip stocks since June 22.

See 3 related tweets

@DynamicWebPaige: 👀 Between this and the impending Qwen release – huge week for open weights models.\n\nQT @Polymark...
@coinbureau: JUST IN: 🇨🇳China's Moonshot AI is planning a Hong Kong IPO within six months after Kimi K3 sent chip...
@markgadala: Wow.. Kimi is currently “sold out”

I was testing out K3 and I ran out of credits so I went to upgra...

19. Teknium (Group Score: 77.3 | Individual: 30.2)

Cluster: 3 tweets | Engagement: 330 (Avg: 532) | Type: Tech

Some of the latest work on the desktop GUI\n\nQT @imbabybrooklyn: Hermes Agent supports multisession tabs and a flexible layout system. Tabs w/in panes, or put anything in your own pane.

Hermes can also generate its own UI elements w/in your workspace if you ask it. More on that later.

@NousResearch @Teknium https://t.co/tG0sLMd3F0

See 2 related tweets

@ivanfioravanti: Hermes Desktop getting better release after release!\n\nQT @imbabybrooklyn: Hermes Agent supports mu...
@Teknium: RT @tonbistudio: Hermes Agent Desktop App is getting some major UI upgrades!

In this short video, ...

20. KobeissiLetter (Group Score: 76.1 | Individual: 24.2)

Cluster: 4 tweets | Engagement: 1854 (Avg: 3264) | Type: Tech

Competition between US and Chinese AI models is heating up:

The proportion of tokens used by US firms that run through Chinese AI models on OpenRouter is up to a record ~58%.

OpenRouter is a platform that allows developers and companies to access and compare AI models from multiple providers, making it a useful gauge of real-world AI model adoption.

This percentage has TRIPLED since mid-January, overtaking American peers on the platform for the first time in March.

The token share of Chinese models briefly rose as high as 63% in the first week of July.

By comparison, Chinese models accounted for less than 10% of usage at the start of 2025, while US models accounted for ~80%.

DeepSeek has emerged as the most popular choice among American firms in recent months.

Chinese AI models are rapidly gaining market share.

See 3 related tweets

@theinformation: OpenRouter, which helps app developers access hundreds of AI models, has discussed a potential sale ...
@WhaleInsider: JUST IN: 🇺🇸🇨🇳 U.S. firms’ token usage on OpenRouter hits record 58% through Chinese AI models...
@EntelligenceAI: 🚨 Chinese AI models have officially overtaken American models on OpenRouter.

Share of global token ...