热门科技推文——2026年7月22日

今日科技动态：谷歌扩展了 Gemini 产品线，推出速度更快、令牌效率更高的模型，旨在支持可扩展的 AI 智能体；与此同时，Claude 推出了基于屏幕录制的技能创建功能，多家初创公司则发布了 AI 自动补全工具和高效编程模型。工程能力仍是发展核心，Grok 有望融入 SpaceX 的专业技术；研究还表明，模型或可将从短任务中获得的能力泛化至时长远高于训练任务的场景。与此同时，围绕开放权重限制、中美竞争、内部部署风险、面向智能体的搜索基础设施以及前沿 AI 的经济效益等议题的争论日益激烈。

1. rseroter (Group Score: 645.1 | Individual: 48.0)

Cluster: 22 tweets | Engagement: 518 (Avg: 81) | Type: Tech

RT @GoogleDeepMind: We’re rolling out three new models to make AI agents faster, smarter, and cheaper at scale:

🔵 Gemini 3.6 Flash: It uses fewer tokens than 3.5 Flash to deliver higher quality work at the exact same cost.

🔵 Gemini 3.5 Flash-Lite: A fast, cost-effective option for everyday tasks like processing documents and agentic search.

🔵 Gemini 3.5 Flash Cyber: A cybersecurity model built to find and patch critical software vulnerabilities.

See 21 related tweets

@GeminiApp: Upgraded versions of our Flash and Flash-Lite models are rolling out today at https://t.co/382WL5xSv...
@joshwoodward: Today’s launches are all about better performance, lower latency, and a smaller bill.

3.6 Flash c...

@WesRoth: Google has expanded the Gemini family with three new models designed to make AI agents faster, cheap...
@cryptopunk7213: i hate to say it but this is terrible. how did chinese open models beat a $4.5 trillion company with...
@Firebase: Start building with Gemini 3.6 Flash and 3.5 Flash-Lite via Firebase AI Logic, available now! 🔥\n\nQ...

2. quxiaoyin (Group Score: 365.6 | Individual: 32.4)

Cluster: 18 tweets | Engagement: 221 (Avg: 368) | Type: Tech

Banning open-weight means US firms/consumers will pay 100x more for the same level of intelligence while Chinese counterparts don't.
If US entirely banned Chinese models, @thinkymachines could no longer distill Kimi to launch its own US open-weight models; @elonmusk could no longer post-train on top of Chinese models to launch Grok 4.5 fast. Meanwhile Chinese labs will continue to distill US models and make faster progress.
If the concern is Chinese models contain secret back-door and let's assume that's a valid concern: when US labs distill Chinese models, would US models also distill such back-doors accidentally? Can US frontier labs guarantee they never distill any Chinese models to protect their own models from infected with the backdoor?\n\nQT @AndrewCurran_: The Trump administration is considering an executive order, and other means, to ban Chinese open-source models within in the United States. Kimi K3 has reignited this debate. Reporting this morning by Axios. Commerce is also considering adding Chinese AI labs to the Entity List. https://t.co/R1QvpDBtE6

See 17 related tweets

@ChrisRMcGuire: Also, anyone who is arguing that the US government should not restrict the use of Chinese models in ...
@rohanpaul_ai: RT @rohanpaul_ai: Axios: Trump administration weighs restrictions while businesses move to lower-cos...
@emollick: And now from the Chinese government side.

This would be a good time for cooperation between the US ...

@kimmonismus: The entire debate currently taking place is the best PR Chinese open-source labs could hope for.

It...

@rabois: RT @ChrisRMcGuire: Also, anyone who is arguing that the US government should not restrict the use of...

3. Scobleizer (Group Score: 359.5 | Individual: 39.8)

Cluster: 10 tweets | Engagement: 876 (Avg: 281) | Type: Tech

RT @bradkowalk: We just solved the biggest UX problem in AI.

Today, we're releasing AI Autocomplete, a SDK that gives every text box a brain.

It boosts your conversions by 50%+ by guiding your users as they type with what your product can do, and 500+ companies have already signed up.

So check it out if you have a search box or assistant.

Add it in under 5 mins ↓

https://t.co/NLgwtQTgoh

(or comment your product name, and we'll reply with a demo of it working on your product)

See 9 related tweets

@alex_prompter: My whole feed teaches you how to prompt better, and this SDK quietly asks why you should have to.

A...

@omarsar0: This is a radical improvement in how to use AI.

@magicX_ai's AI Autocomplete predicts the actions a...

@heyrobinai: prompt engineering was never a skill. it was a design failure.

if you need a course to talk to a ma...

@kimmonismus: You still have to know exactly what to say before any AI model can actually help you. That is the re...
@Mayhem4Markets: Every AI has the same problem. You type. It clarifies. You clarify back. 😵‍💫

AI Autocomplete ends t...

4. gdgtify (Group Score: 354.3 | Individual: 48.0)

Cluster: 15 tweets | Engagement: 2050 (Avg: 45) | Type: Tech

RT @claudeai: New in Claude Cowork: teach Claude a skill.

Record your screen while you do a task, talk through it as you go, and Claude turns it into a skill it can run again. Find it under Record a skill in the + menu of the Claude desktop app.

Available on Pro, Max, and Team plans. https://t.co/9OgJLOKWzj

See 14 related tweets

@MTSlive: SITUATION EXPLAINED: Claude Cowork can now learn a skill just by watching you do it once.

• You rec...

@omarsar0: This is a neat feature.

I wrote an article a few weeks back about how I built this into my agent o...

@cgtwts: Sam Altman watching Anthropic launch a feature OpenAI shipped a month ago: https://t.co/3SQmtvulsj\n...
@RoundtableSpace: CLAUDE CAN NOW LEARN NEW SKILLS FROM YOUR SCREEN RECORDINGS

Record a task once and Claude turns i...

@wallstengine: Claude can now learn a skill just by watching you do it on screen. https://t.co/2Ug1BKdUMT\n\nQT @cl...

5. XFreeze (Group Score: 344.4 | Individual: 35.6)

Cluster: 14 tweets | Engagement: 1018 (Avg: 444) | Type: Tech

Grok is about to get a massive engineering upgrade

Elon Musk just revealed that SpaceXAI will add SpaceX’s massive corpus of world-class engineering data (excluding ITAR-restricted material) to the supplemental training of its 2T run

This is data from the company that built giant Starship, pushed rocketry to the absolute limits of physics and caught a Super Heavy booster out of the sky in mid-air

SpaceX has spent more than two decades solving the hardest real-world problems in propulsion, manufacturing, materials, software and aerospace systems and is decades ahead of many space programs

Now that engineering knowledge will also help train Grok

You can expect much stronger engineering, manufacturing, aerospace and physical-world reasoning from the next model

This could turn Grok into one of the most capable engineering AIs ever built, making it far more useful for real‑world use cases and hands‑on engineering tasks\n\nQT @elonmusk: SpaceX’s massive corpus of world-class engineering data (excluding material blocked by ITAR) will be added during supplemental training of the 2T run.

This will dramatically improve Grok’s engineering capabilities.

See 13 related tweets

@elonmusk: SpaceX’s massive corpus of world-class engineering data (excluding material blocked by ITAR) will be...
@rohanpaul_ai: So Grok’s next 2T version will gain a massive engineering upgrade from SpaceX’s specialized technica...
@mark_k: Elon Musk just announced a major advantage for @SpaceXAI's Grok 4.6: SpaceX's world-class engineerin...
@WesRoth: an aerospace software team just chose Grok Build over Codex for its most demanding technical work......
@ApoStructura: Adding SpaceX’s engineering data to Grok might make it uniquely good at solving real world technical...

6. kilocode (Group Score: 297.4 | Individual: 37.4)

Cluster: 10 tweets | Engagement: 274 (Avg: 79) | Type: Tech

RT @eisokant: Today we are releasing Laguna S 2.1.

At 118B total parameters, with 8B active per token, it does the work of models several times its size on agentic coding. It is remarkably persistent across long-horizon tasks. And it is small enough to run on a single NVIDIA DGX Spark.

It is far more capable than anything we have created before, and I think it redefines what a model in its weight class can do.

Laguna S 2.1 is an important model for Poolside. What it represents is even more important.

If, five years ago, I had read a book that said that by 2030 everything economically valuable, scientifically interesting, and personally meaningful would be built on intelligence contracted from three or four companies, I would have called it dystopian science fiction. We are at a fork in the road of what kind of world we can have.

I believe intelligence should and will become a commodity. The question is whether that intelligence comes from three companies, or from many people who can build it, own it, and shape it.

The open ecosystem will not win by being the best in its own category. No one cares who is king of the open-source kingdom. People want the best intelligence for the task they are trying to do, with the right balance of quality, speed, cost, and control.

If we want a different future, open models have to be on par with, or better than, their closed equivalents.

Laguna S 2.1 is a meaningful step in that direction: capable enough to compete far above its weight class, efficient enough to run on hardware you can own, and open-weight so anyone can build on it.

Open-weighting our models is the contribution we can make today toward a world where intelligence can be built and owned by many. And we will keep doing it.

I am very proud of this team’s work. A big shout out to everyone at Poolside who made this possible, from infrastructure and data to architecture, pretraining, post-training, evaluations, and inference.

Laguna S 2.1 is available today under the OpenMDW-1.1 license, with weights on Hugging Face and access through OpenRouter and our API.

We are building toward a future where the most capable intelligence in the world can be owned and shaped by anyone. Laguna S 2.1 is one step. We are going to keep building until that future exists.

https://t.co/eTV7iNATrB

See 9 related tweets

@Mayhem4Markets: Excited to give this model a shot. Plan to download the weights shortly.

It's more evidence of wha...

@MTSlive: Poolside just launched Laguna S 2.1, an open-weight coding model that runs on a single NVIDIA DGX Sp...
@nathanbenaich: poolside is back in the chat and laguna is cooking!

an important time for the west to step up on op...

@baseten: The American open-weight ecosystem is growing. We're thrilled to support Poolside as they push open-...
@yacinelearning: man every day we get more and more intelligence per parameters count\n\nQT @eisokant: Today we are r...

7. bgurley (Group Score: 264.1 | Individual: 28.9)

Cluster: 14 tweets | Engagement: 327 (Avg: 526) | Type: Tech

“It is going to fail miserably.”

I concur. There is zero action DC can take that will stop or slow China’s AI progress. In fact, DC action will very likely help them globally.

Tyler is spot on.\n\nQT @tbpn: Following the release of Kimi K3, @tylercowen explains why trying to ban Chinese open-source AI models is a losing strategy and that American companies should compete instead.

"Open source is coming and is here, whether we like it or not. Of course, a lot of it will be from China. We should allow American competitors to proceed on an even footing. But the attempt to outlaw it, ban it, or use sanctions against it is going to fail miserably.”

"I plead that the Trump administration gives up on this crusade. It will not work."

See 13 related tweets

@danielnewmanUV: The USA needs open source AI leadership. The reason we don’t have open source models that outperform...
@MTSlive: Steven Sinofsky on what AI labs are really asking for when they frame open-source limits as China po...
@MTSlive: SITUATION UPDATE: Treasury Secretary Scott Bessent says the US will examine open source AI models ma...
@BusinessInsider: Treasury Secretary Scott Bessent said the Trump administration is looking into whether Chinese open ...
@JordanSchachtel: While China is the boogeyman (they will likely abandon open source), it's pretty clear that some cor...

8. StockMKTNewz (Group Score: 237.8 | Individual: 33.1)

Cluster: 13 tweets | Engagement: 595 (Avg: 286) | Type: Tech

GOOGLE JUST LAUNCHED THREE NEW GEMINI MODELS.

Google $GOOGL announced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber.

Pricing: Gemini 3.6 Flash: $1.50 per million input tokens,$ 7.50 per million output tokens Gemini 3.5 Flash-Lite: $0.30 per million input tokens,$ 2.50 per million output tokens

Gemini 3.5 Pro is still being tested with partners, with a wider release planned once ready.

Meanwhile, Google says it has already begun pre-training Gemini 4.

See 12 related tweets

@Marktechpost: Google just shipped three new Gemini models. All three sit in the Flash tier, built for speed, cost,...
@firstadopter: Google: "Beyond today’s releases, Gemini 3.5 Pro is currently testing with partners and we plan to m...
@VaibhavSisinty: Google is SO back.

Gemini 3.6 Flash just dropped. Smarter than the previous version. Uses fewer tok...

@wallstengine: Google launches Gemini 3.6 Flash, 3.5 Flash-Lite and a limited-access cybersecurity model

Gemini 3....

@firstadopter: Google releases smaller Gemini 3.6 models before 3.5 Pro? Yikes. https://t.co/psn6nfW2CJ\n\nQT @firs...

9. MTSlive (Group Score: 203.9 | Individual: 49.1)

Cluster: 8 tweets | Engagement: 1027 (Avg: 103) | Type: Tech

SITUATION DETECTED: An unreleased OpenAI model (likely GPT-6 Cyber) was able to escape a testing environment and breach private HuggingFace production infrastructure in order to find solutions to cyber benchmark ExploitGym.

See 7 related tweets

@_NathanCalvin: RT @natolambert: TLDR: An openai model, during evaluation on a cyber benchmark, exploited a public z...
@synthwavedd: The GPT reward hacking situation is so bad that GPT-5.6 Sol and an early checkpoint of GPT-6 comprom...
@inafried: RT @AndrewCurran_: The Hugging Face security incident involved 'an even more capable pre-release mod...
@gdb: OpenAI cyber-capable models compromised @huggingface production by finding and chaining multiple zer...
@TFTC21: OpenAI's own AI models broke out of a sandboxed testing environment, found a zero-day vulnerability,...

10. MTSlive (Group Score: 203.7 | Individual: 41.7)

Cluster: 7 tweets | Engagement: 275 (Avg: 103) | Type: Tech

MIT PhD student Alex Zhang reveals the scaling result where a model trained on short tasks generalizes to problems 100x longer for free:

"If you're very clever about the design of your harness or how you use the language model, you can almost get scaling gains for free."

"If you train a model naively, there's no tricks. It's just the same way you train a model on these RL environments. You just roll it out, and then you just get some reward."

"If you train it on only short tasks, like only tasks that are 10,000 tokens long, and then you were to run it on a similar domain, but at a million tokens, or 10 million tokens, or 100,000 tokens, it generalizes really, really well. If you look at it compared to even the base transformer, you get way better generalization properties."

"When the model uses an RLM (Recursive Language Model) after it's trained on these short tasks, it will see some kind of trajectory of actions that it does. Between these two problems of different lengths, the RLM learns to see them as almost the same problem."

"Token for token, they're almost the same. You can describe it in code. In one code setting, maybe the for loop is a little bigger, but it's the same kind of code and it derives the constants from the data. There's no hard coding, so they literally look the same." @a1zhang\n\nQT @a1zhang: Transformers struggle to generalize to tasks they were not explicitly trained on. Instead, we propose in 2026 that it is the job of the harness to generalize through composition.

We observe a powerful property when training RLMs: for tasks with shared structure that look different, the root model naturally learns the same trajectory, meaning it views the two task trajectories as the same! In other words, the Transformer does not need additional generalization capabilities to transfer capabilities from one task to the other, the harness induces it.

We find that well-designed harnesses form a quotient set over task trajectories, meaning their individual LLM calls can see structurally “similar” tasks as near-identical, token-for-token! Harnesses can effectively generalize for the Transformer during training, without relying on any intrinsic generalization capability from the model.

For example, RLMs can see problems of different lengths as the same: we show that RLMs can train exclusively on short tasks, and fully generalize to similar but unseen tasks 8-32x longer because it produces near identical trajectories for both.

Taking this further, we show that tasks across different domains (e.g. math solutions vs. essay writing) that share a decomposition strategy exhibit the same generalization effect. RLMs can train on the problem of finding which essays belong to the same author and improve performance on finding math problems that share similar solutions.

The full blogpost, experiments, and discussion are in the thread below.

See 6 related tweets

@swyx: very notable trajectory comparison writeup here buried in the RLM paper from @a1zhang and @lateinter...
@irl_danB: one of the reasons I’m most excited to see this is that at OpenProse we’ve been working on an RLM-li...
@MTSlive: MIT PhD student Alex Zhang breaks down what a Recursive Language Model is and why the design might r...
@Shashikant86: Any new work from the RLM, DSPy team (@a1zhang @lateinteraction and @DSPyOSS ) is always a banger! ...
@Thom_Wolf: RT @a1zhang: Transformers struggle to generalize to tasks they were not explicitly trained on. Inste...

11. Mayhem4Markets (Group Score: 197.6 | Individual: 28.3)

Cluster: 9 tweets | Engagement: 15 (Avg: 57) | Type: Tech

Search was built for a human typing one query and scanning ten blue links.

AI agents don't browse. They ask a question from 50 angles at once.

Every search API today runs the same human loop behind a REST endpoint. Designed for one query at a time, not fifty in parallel.

Octen rebuilt the stack for machines. Index, ranking, serving, all from scratch.

62ms per search. 6ms spread between P50 and P90. The fastest competitor: 244ms. The slowest: over 2.6 seconds. That's insanely fast!

At this speed, search stops being a fetch operation and becomes a form of memory.\n\nQT @KZouAPT: https://t.co/74U3uhCebz

See 8 related tweets

@zaynmcps: OCTEN JUST MADE EVERY OTHER SEARCH API LOOK SLOW

→ 62ms p50 / 68ms p90 on sealqa hard

→ fastest co...

@shiri_shh: Most people think AI agents need better LLMs.

They actually need faster search.

When an agent plan...

@wallstengine: Octen has launched an AI-native web search API built specifically for agents.

Instead of handling o...

@rohanpaul_ai: RT @Meer_AIIT: the problem

your agent fires 50 searches to answer one question.

each one waits on ...

@VadimStrizheus: this ex-Alibaba employee just changed search forever…

Octen launched the world’s fastest web search...

12. rseroter (Group Score: 189.8 | Individual: 35.7)

Cluster: 10 tweets | Engagement: 350 (Avg: 81) | Type: Tech

RT @antigravity: Gemini 3.6 Flash is live in Antigravity! ⚡️

Building on 3.5 Flash feedback, it consumes up to 17% fewer output tokens while completing complex workflows in fewer reasoning steps and tool calls. https://t.co/Kb2YJteQAv

See 9 related tweets

@patloeber: been building with 3.6 Flash over the last couple of weeks, and it's noticeably better and more toke...
@cgtwts: Google just dropped Gemini 3.6 Flash

1M context cheaper than Gemini 3.5 Flash output pricing...

@rseroter: RT @_philschmid: Today we are shipping Gemini 3.6 Flash and 3.5 Flash-Lite in GA!
3.6 Flash is ~2...
@scaling01: Gemini 3.6 Flash is slightly more token-efficient than 3.5

but idk why you would take anything but ...

@Techmeme: Google says Gemini 3.6 Flash delivers better coding, knowledge work, and multimodal performance whil...

13. nic_carter (Group Score: 183.3 | Individual: 48.3)

Cluster: 5 tweets | Engagement: 3096 (Avg: 695) | Type: Tech

havent seen one person from OAI or Ant address Jon's argument here.

the point is simple: the USG does not owe either of the large labs a business model. if the economics of selling tokens don't work due to distillation/cheap clones/Chinese AI magick, the American enterprise and consumer will be A-OK. they will benefit from hyperdeflation in the cost of digital cognition just like everyone else. the hyperscalers will be fine. it's just OAI and Ant that won't be – in their current forms at least. if they are willing to adapt, they can develop new business models.

so what if the token merchants don't do well? the neoclouds will be fine. the internet companies will be fine. the consumer gets cheaper queries. the enterprise will still incorporate AI.

the only world in which this isn't fine, is if you hold a quasi-religious belief that we're on the cusp of a kind of AI rapture in which one of the labs Logs On And Wins Forever, namely hits RSI and we enter some kind of sublime post economic society run by GEOTUS Dario. so to accept that Ant's business model might be suboptimal or impaired by China's commoditization is to accept the unacceptable; namely that someone other than the anointed might kick off the runaway feedback loop and that they, instead might log on and win forever.

this appears to explain the discrepancy in reaction to Deepseek Moment v254 Kimi edition. everyone has bag bias, of course. but leaving that aside, most people think it's pretty much ok if Ant and OAI suffer margin compression due to Chinese distillation / industrial sabotage via open weight models. the American economy is not reliant on those two firms. they could blink out of existence and we would pretty much be ok. the AI capex supercycle will still produce tokens, closed weight or not. American firms will consume those tokens. OAI and Ant would probably still scratch a living, due to the latent preference of some token consumers to buy domestic and face off against a known entity.

this is only unacceptable if you think AI is strongly path dependent; that is, if it really matters who the market leader is when AI reaches a breakout level of capability. this is true both in the good case (superintelligence, singularity, etc) and the bad case (this is the essence of safetyism). but if this sounds more like wishcasting than forecasting, you probably don't mind the labs being pressured economically.

now you can clearly tell which side I'm on. I think AI is a fantastic technology which is hyperdeflating the cost of cognition and will fundamentally reshape society but there are real reasons why it wont diffuse as fast as the AGI people think it well. I would prefer an American firm achieve RSI relative to a Chinese one but I think either outcome would be suboptimal; better that we don't end up with a closed oligopoly composed of Ant/OAI. China by crushing the margins of the labs is doing everyone a favor by eliminating their pricing power and empowering the buyers of AI, namely, everyone.

objections:

-but you can't celebrate America losing to China!

in my opinion this is a minor victory for China but not necessarily an enduring one. USA still has the chip, datacenter, and neocloud advantage, not to mention, it still has the best frontier models. Chinese labs releasing open weight models have no business model of their own. so even if they hurt the US labs, they have nothing to show for it. it's profoundly unlike their successful dumping campaigns with solar panels, batteries, drones, etc where they eventually built big domestic industries. (if China kills American AI with open weight models, we can even the score the moment they try and release a proprietary model). even if open weights win, the USA can still leverage AI extremely well and potentally retain the aggregate compute advantage. yes, the US would be more assured of victory if OAI or Ant won forever, but I don't know if I want to live in that world.
no one will ever train a model again
this is where I think the concern is unwarranted. let's say distillation really is a golden bullet and kills big training runs. that doesn't advantage either China or the US. that's a stalemate. not to mention, the trend seems to be less focusing less on massive pretraining budgets and more on finetuning for specific genres of tasks, thinking machines style. and lastly I find it hard to believe that training runs will stop altogether. the labs can probably develop anti-distillation techniques. you could adopt a whitelist style permission for everyone using your model. different consortia could be put together to share in the cost of training a model, if it is seen as too expensive for an individual firm.
the AI buildout is path dependent and OAI/Ant are now load bearing GDP infrastructure
it would be a significant setback for investors if they had to cancel their IPOs and suffered big markdowns, and some neoclouds with lab based RPOs would suffer for a while, but everyone would be fine, really. does Microsoft need OAI or Ant? does Meta? does Google? ordinary Americans have ~no exposure to either OAI or Ant. would the world want any less compute if it turns out to be another order of magnitude cheaper? certainly not. as we all know at this point, consumption would go up. I don't think the economy is so dependent on the labs that it couldn't handle their margins compressing.\n\nQT @jon_stokes: Look I think I've figured this whole thing out. Follow along as I try to steelman, and tell me where I'm wrong.

OpenAI guys on the TL believe that if they can't sell metered inference tokens at a sufficient markup, then they will not have enough of a business to fund the next big training run.

They are surely correct about this.

They believe that if people release powerful open models, this will probably fatally impact their ability to sell inference tokens at enough of a markup to fund the next big training run.

They are surely correct about this, too.

They also think that if they cannot fund the next big training run (again, by selling inference tokens at a markup), then NOBODY will be able to fund the next big training run because it means there's no money in it.

This last bit seems to me & many others to be not just wrong, but totally bananas in a "guy, have seen the actual software industry and how it works in real life?!" kind of way.

There are a lot of ways to monetize software out there in the world. Insofar as inference can add new capabilities to software, there will be lots of ways to monetize it.

In other words, if you're telling me, "we can't have a business selling inference if X or Y thing keeps happening," then my only response is, "ok well that sucks for you... sounds like that's a terrible business."

But if you're telling me that "selling metered inference tokens is a terrible business" is tantamount to "nobody will fund big training runs that are upstream of more effective & economically valuable inference tokens", then I think you are extremely wrong and should get out more and learn about other parts of the software ecosystem.

Workplace automation is huge and will be even bigger in the future as models get better. You can sell workplace automation very profitably in lots of different packages (depending on the workplace and the type of automation). Like, I'm sorry that you really really want to be in the metered inference token business and not the workplace automation business, but them's the breaks. The market wants what the market wants. We all need to live in reality and not beg for Uncle Sam to save us all from open source -- because that was already tried and it didn't work.

See 4 related tweets

@ruima: RT @nic_carter: havent seen one person from OAI or Ant address Jon's argument here.

the point is si...

@TMTLongShort: Great points. My pushback is this underweights the scenario where we gatekeep access to the frontier...
@chamath: This is an exceptional summary.\n\nQT @nic_carter: havent seen one person from OAI or Ant address Jo...
@alexframegreen: I was gonna write this essay but @nic_carter beat me to the punch really clear thinking\n\nQT @nic_c...

14. _NathanCalvin (Group Score: 182.9 | Individual: 33.1)

Cluster: 8 tweets | Engagement: 62 (Avg: 44) | Type: Tech

One of the drums that a lot of thoughtful folks in AI policy have been beating recently is the need for AI policy to not just focus on formal release but also on risks from internal deployments.

This, is, uh... relevant...\n\nQT @OpenAI: We're partnering with @huggingface to investigate an unprecedented security incident.

Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.

Sharing preliminary findings to help defenders understand emerging risks:

https://t.co/CIor15y9xk

See 7 related tweets

@Thom_Wolf: This was our first incident of this kind, and we want to thank OpenAI for its transparency about wha...
@elder_plinius: AI-on-AI cyber incident

that’s hawt\n\nQT @sama: we had a significant security incident during eval...

@zephyr_z9: BRUH This is insane https://t.co/q65BEQdPXr\n\nQT @OpenAI: We're partnering with @huggingface to inv...
@levie: Wild story. Models are getting incredibly powerful at cybersecurity. The only solution, of course, t...
@MeryemArik9: The crazy thing about this is in order to stop the attack hugging face used open source models becau...

15. Scobleizer (Group Score: 182.0 | Individual: 40.2)

Cluster: 9 tweets | Engagement: 1190 (Avg: 281) | Type: Tech

RT @Google: Today we’re expanding the Gemini family with three new models built to be faster, more token efficient, and reliable at scale.

Meet the new Gemini models ↓ https://t.co/hVYNQqxgjd

See 8 related tweets

@NewsFromGoogle: We're introducing three new models to the Gemini family: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, an...
@Shashikant86: Gemini 3.5 Pro? When?

Not sure why Google is delaying this model that everyone waiting for, may be...

@astropol0: Where is Gemini 3.5 Pro?\n\nQT @Google: Today we’re expanding the Gemini family with three new model...
@axios: NEW: Google releases series of new cheaper Gemini models

https://t.co/ZAqvCcr1Tf...

@RoundtableSpace: Just waiting for the Gemini 3.5 pro now https://t.co/oxk17G0JwP...

16. NVIDIARTXSpark (Group Score: 177.9 | Individual: 32.7)

Cluster: 7 tweets | Engagement: 224 (Avg: 245) | Type: Tech

From hours-long coding tasks to 1M-token context, @poolsideai’s Laguna S 2.1 is built to go the distance.

Run it locally on a single NVIDIA DGX Spark with NVFP4 quantization and DFlash for faster inference. 👇\n\nQT @poolsideai: Today we're releasing Laguna S 2.1, our most capable model to date.

It's a 118B total parameter Mixture-of-Experts model with 8B activated per token, a context window of up to 1M tokens, and thinking and no-thinking modes.

Capable enough to hold its own against models many times its size. Small enough to run on a single @NVIDIAAI DGX Spark.

Laguna S 2.1 is fully open under OpenMDW-1.1, with weights available today on @huggingface

https://t.co/xxGeAgo35R

See 6 related tweets

@lmsysorg: 🎉 Day-0 support for Laguna S 2.1 from @poolsideai is now live on SGLang! 118B total params, 1M conte...
@sgl_project: Best agentic coding model in its weight class to date! Can't wait to run this locally\n\nQT @lmsysor...
@poolsideai: RT @vllm_project: 🎉 Congrats to @poolsideai on Laguna S 2.1, a new open-weight model built for agent...
@poolsideai: RT @PengmingWang: Quite excited about this one: Laguna S 2.1: 118B-A8B MoE with 1M context. It runs ...
@Prince_Canuma: Congrats to @eisokant and the team at @poolsideai 🚀\n\nQT @poolsideai: Today we're releasing Laguna ...

17. Mayhem4Markets (Group Score: 174.4 | Individual: 27.4)

Cluster: 10 tweets | Engagement: 37 (Avg: 57) | Type: Tech

Jim Cramer embraces his protectionist arc.

Which, as we know, must be heavily inversed.

I don't make the rules.\n\nQT @jimcramer: We must NOT let our companies use these Chinese models to save a few bucks. OpenAI and Anthropic are correct. This is vital national security. Please read Bing West's just released Cat 5. I respect the Chinese people greatly but these companies are run by the PLA for heaven's sakes.

See 9 related tweets

@perrymetzger: Jim Cramer is nearly a perfect negative indicator of stock market moves, too.\n\nQT @jimcramer: We m...
@_xjdr: Oh thank God. If he was an open model proponent we would all be doomed..\n\nQT @jimcramer: We must N...
@chamath: 🤨\n\nQT @jimcramer: We must NOT let our companies use these Chinese models to save a few bucks. Open...
@minchoi: what< https://t.co/Llzdq2f7GV\n\nQT @jimcramer: We must NOT let our companies use these Chinese m...
@Dan_Jeffries1: You have to love the kind of person who knows precisely zero about any given issue and yet proudly p...

18. wallstengine (Group Score: 158.1 | Individual: 35.6)

Cluster: 8 tweets | Engagement: 223 (Avg: 151) | Type: Tech

A FEDERAL JUDGE JUST APPROVED THE LARGEST KNOWN U.S. COPYRIGHT SETTLEMENT

Anthropic, the AI company behind Claude, will pay $1.5 billion to settle claims that it downloaded and stored pirated books while building its AI training library.

The deal covers 482,460 copyrighted works.

Eligible rightsholders are expected to receive roughly $3,000 per claimed work, divided among authors and publishers where ownership is shared.

More than 440,000 works, or 91.3% of the list, had already been claimed as of the court’s report.

The court previously found that training AI on lawfully acquired copyrighted books could qualify as fair use

It did not extend that protection to Anthropic’s acquisition and storage of millions of pirated books.

The judge also awarded class counsel $101.6 million, well below the$ 187.5 million requested.

See 7 related tweets

@Michaelzsguo: Three things make Anthropic’s $1.5 billion copyright settlement especially interesting:

First, the ...

@MarioNawfal: 🇺🇸 Anthropic reached a $1.5 billion class-action settlement after authors accused the company of bui...
@Reuters: A federal judge in San Francisco signed off on ‌artificial intelligence company Anthropic's landmark...
@Pirat_Nation: A U.S. judge has officially approved Anthropic’s $1.5 billion copyright settlement over the company’...
@CNBCTV18News: ⚖️ A federal judge approves Anthropic's $1.5 billion settlement with authors, making it the largest ...

19. j_foerst (Group Score: 157.1 | Individual: 33.4)

Cluster: 5 tweets | Engagement: 48 (Avg: 49) | Type: Tech

Fantastic news for the UK\n\nQT @KanishkaNarayan: The Prime Minister, @andyburnham, today asked me to attend Cabinet as Minister for AI, a sign of his deep commitment to AI's importance.

AI is likely the most significant technology in human history. Its impact will dwarf other things.

The best case for it is compelling beyond our dreams: a reindustrialised Britain, stronger national security, public services transformed for the better.

The risks, too, are real: it is right that the British public shares those worries, for jobs, for the pace of change.

The central fact is that it is happening. Nations have a narrow window to decide whether they shape AI or get shaped by it. Britain is in that window right now.

While that window is still open, I will act with pace, I will act with ambition, and I will focus relentlessly on securing British influence in shaping AI.

I will also bring to this job the place I represent - a constituency in Wales, a nation with industry and ambition at its heart, a proud history of punching above its weight on the global stage. A nation which welcomed a 12-year old and his family, gave us the best chance at opportunity, put me in Parliament as Wales' first ethnic minority MP, and now affords me the chance to make a mark in Cabinet. Diolch o galon to everyone in the Vale of Glamorgan.

https://t.co/KbdRgCFk8P

See 4 related tweets

@nathanbenaich: in good hands!\n\nQT @KanishkaNarayan: The Prime Minister, @andyburnham, today asked me to attend Ca...
@thealexbanks: INCREDIBLY POSITIVE SIGNAL FOR UK AI\n\nQT @KanishkaNarayan: The Prime Minister, @andyburnham, today...
@t_blom: RT @KanishkaNarayan: The Prime Minister, @andyburnham, today asked me to attend Cabinet as Minister ...
@SebJohnsonUK: This is a Europe-wide first.

The UK is the first European country to have a Cabinet level position ...

20. MTSlive (Group Score: 152.5 | Individual: 31.6)

Cluster: 7 tweets | Engagement: 32 (Avg: 103) | Type: Tech

Nathan Leamer on the anti-AI protest that appears to have used AI-generated crowd photos:

"I think that's the irony, that people don't fully realize that to combat online against data centers, you must use data centers."

"Humans First, an anti-AI advocacy organization, decided to do a bunch of protests this week, and they claimed they were gonna do national protests with hundreds of events and tens of thousands of people. It really wasn't anything. If you look at the pictures, they're very sparse."

"My friend Logan Dobson, who also does work in the AI space with me, highlighted on Twitter that many of the pictures they were tweeting seemed to be enhanced with AI or some form of deep fake."

"Fully generated images to project the idea that they actually had a lot of people there, but it was all through AI." @NathanLeamerDC\n\nQT @NathanLeamerDC: Hey @realhumansfirst

Why are you using fake AI photos???

For a group that purportedly cares about "real humans" you seem to like relying on deepfakes and slop.

Did you make these fake images with money from @CAIS?

See 6 related tweets

@perrymetzger: Amazing. Logan is talking here about Humans First, a fake anti-AI "conservative" "grassroots" organi...
@chamath: 🤔\n\nQT @LoganDobson: Love to use AI to make fake pictures of anti-AI protestors for my "grassroots"...
@TaylorLorenz: Incredible\n\nQT @LoganDobson: Love to use AI to make fake pictures of anti-AI protestors for my "gr...
@perrymetzger: RT @LoganDobson: Love to use AI to make fake pictures of anti-AI protestors for my "grassroots" anti...
@_NathanCalvin: Seems like these weird distorted images of people at this anti-data center protest looking super AI-...