热门科技推文 - 2026年7月31日

今日科技领域，人工智能的经济效益成为焦点。OpenAI下调模型价格，并重点介绍可通过自我优化降低服务成本的系统；与此同时，市场对前沿人工智能实验室估值及人工智能主题基金的质疑日益加剧。基础设施需求激增推动微软和三星创下业绩纪录；此外，谷歌支持的Anthropic计划在得克萨斯州建设园区，欧洲也在推进人工智能“千兆工厂”。工程技术方面的进展包括多模态机器人和可在本地运行的开放模型。初创企业正迎来更多机遇，但数据来源、能源需求和融资渠道等问题也引发了越来越多的担忧。

1. firstadopter (Group Score: 480.9 | Individual: 44.5)

Cluster: 20 tweets | Engagement: 1012 (Avg: 179) | Type: Tech

Welp. The news has finally hit the tape.

FT: "Situational Awareness, the $20bn hedge fund founded by former OpenAI employee Leopold Aschenbrenner, has sought to raise fresh capital from investors after suffering heavy losses during the recent rout in AI stocks."

"Several people familiar with the matter said that Situational Awareness had used borrowing to magnify its returns, a popular hedge fund strategy that can also amplify losses in a downturn."

"The firm was up 439 per cent on a net basis for the year as of the end of June"

"Aschenbrenner pointed in the letter to the prospect of major AI developments in the second half of the year"

Probably RSI https://t.co/R8uSvznwNO

See 19 related tweets

@firstadopter: David Faber on CNBC just said Situational Awareness got to $24 billion AUM at near highs. "It is lev...
@exec_sum: BREAKING: Situational Awareness, the $20B hedge fund of ex-OpenAI employee Leopold Aschenbrenner, is...
@firstadopter: FT: "Situational Awareness rapidly sold a large portion of its $16bn public equity holdings to Citad...
@rohanpaul_ai: FT: Leopold Aschenbrenner’s (Ex OpenAI researcher) $20B hedge fund, Situational Awareness, is seekin...
@FT: FT Exclusive: Ken Griffin’s Citadel swooped in to buy a large portion of hedge fund Situational Awar...

2. Dimillian (Group Score: 410.5 | Individual: 45.4)

Cluster: 20 tweets | Engagement: 1774 (Avg: 88) | Type: Tech

RT @OpenAI: We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

See 19 related tweets

@tenobrus: gemini team on suicide watch rn\n\nQT @OpenAI: We are committed to pushing the model frontier across...
@kunchenguid: gpt-5.6-luna price reduced by 80% (!!!)

it was already cheap - this will make it feel like almost ...

@VadimStrizheus: OpenAI just killed the cheap model market.

GPT-5.6 Luna now beats Gemini Flash and every Chinese mo...

@trevin: By surfing your subscription plans, you can make them go much much further. I've been using Fable t...
@cgtwts: openai just killed gemini and several chinese open-source models with these price cuts. https://t.co...

3. andrewho03 (Group Score: 298.0 | Individual: 54.7)

Cluster: 8 tweets | Engagement: 1717 (Avg: 292) | Type: Tech

I'm actually fairly bearish on frontier lab valuations. I've never seen the reasons articulated to my satisfaction, so before I go to sleep, I wanted to quickly jot down my thinking here.

The basic issue is that the labs are highly unprofitable. This may seem like a simple point, but private market valuations can be relatively irrational; however, like with $SPCX, post-IPO pricing will likely be much more punishing, especially as the standard 6-month lockup period expires and selling pressure intensifies.

Many people claim that the labs have high margins. Yet even with high margins, a valuation of $1T would be justified only if the labs were doing nothing aside from serving inference (thus reducing costs only to those relevant to inference) and posting annual revenue numbers in the$ 100-200 billion range assuming ~80% gross margin and a 20x earnings multiple.

This assumption is obviously not true, because the frontier labs have to continually spend money training the next generation of models. This is because of market competition from runner-up firms. For example, if OpenAI had paused model development last year, there would no longer be any point in paying GPT-5 API prices when you can just use Qwen or Kimi instead for much cheaper. Thus, the labs are forced to invest ever-increasing amounts of money in model training, in a way such that at any given point of time, the amount you're forced to invest in the next model is dramatically higher than the amount of money you're actually making, because even if your revenue goes up with higher model capabilities, so do your future training costs. This is a profoundly punishing dynamic which severely penalizes frontrunners.

(There is also a related subpoint where frontier labs claim they can distill their leading models to win out at lower intelligence levels as well. This makes no sense because the revenue numbers involved are far too low when taking into consideration the rather low margin of such inference.)

Frontier lab valuations appear largely to be based on the assumption that as you scale up, the capabilities which emerge will be sufficiently general and profound that we'll see explosive growth (https://t.co/RqmkltVpM3) from things akin to AI agents starting and autonomously managing entire companies of subagents. But it's not clear to me that this is the case; indeed, as I mentioned in my previous post (https://t.co/3URAcJ4XkJ), I believe that capabilities growth will be slower, spikier, and more data-limited than people currently assume. It may be the case that eventually we will see explosive growth of this nature with full automation of the economy, but at the very least my viewpoint implies much longer (multi-decade) timelines until we reach this point. It is not clear to me that the frontier labs will be able to operate unprofitably for so long, although I suppose maybe this foreshadows some sort of inevitable nationalization.

I also want to make a broader point about technological diffusion. The reason why technological diffusion is slow isn't just because, e.g., old people take a long time to learn how to use technology (although this is of course a contributing factor to some degree). In my view, it's because when a new, revolutionary technology comes along, the ways to incorporate that technology into subsequent developments are not always obvious, and in fact they cannot necessarily be arrived at through the application of pure reason. If they could be, then perhaps frontier models, at a certain point, would have a perfect understanding of how the LLM application layer should be developed, and they would then autonomously code, deploy, and sell such a layer.

But it seems more plausible to me that this diffusion is limited moreso by the hard problem of economic calculation--that is to say, the Hayekian notion through which the price system gradually promotes efficient allocation of resources and which cannot be simulated through central planning--and that even if we froze current capability levels at today's levels, it would take well over two decades to fully integrate in LLMs into our lives. Such a view is consequently rather bearish for the continued profitability of labs as it reduces their prospects for finding, say, something else comparable in profitability to coding agents, which seems to have been a somewhat lucky discovery by Anthropic to begin with. That is to say, even if you spam FDEs you aren't necessarily going to be able to just figure out the "correct" product shapes fast enough.

Overall, I don't think that people have clearly reasoned through their mental models for why lab equity should be worth as much as it currently is, and that if you actually bother to write down such a model, you may not arrive at the conclusion that you want to arrive at. This isn't to say that I don't expect AI to experience a huge (industry-wide) boom in the coming decades, but just that I'm not entirely sure I would buy OpenAI or Anthropic stock at latest valuations if I were given the opportunity to do so.

Of course, as an ex-lab employee, arguably this is talking against my own book; I should really be giving people more reasons to be bullish. But in the end, my influence is so small that it doesn't make a difference, so why not have some fun?\n\nQT @dwarkesh_sp: New blog post on what would be true about the world if trendline continues and leading lab hits $1T in revenue by the end of next year.

In other words, why compute might get 10x+ more expensive in coming years

https://t.co/7PvniDC2jN https://t.co/qJ1yhSkdyd

See 7 related tweets

@willdepue: i think this would be a great take if it weren’t for AGI. capabilities will not slow down, RSI is le...
@andrewho03: Also, for my friends at the frontier labs, I would strongly recommend taking liquidity if you're eli...
@natolambert: I'm somewhere in between this post and RSI, where the current valuations hold up, but the frontier l...
@andrewho03: Also important to note here—I believe that to be a maximally effective employee, you have to think s...
@ShanuMathew93: Andrew on the BEAR case. Now the obvious cavaet is Andrew may be motivated to talk the book around d...

4. MelvinInvests (Group Score: 282.9 | Individual: 54.8)

Cluster: 9 tweets | Engagement: 395 (Avg: 61) | Type: Tech

This is WILD!

Citadel manipulated the entire market along with Leopold's portfolio and now Ken Griffin's shop is scooping up the wreckage.

Citadel previously called for the Fed to hike rates, a signal that helped drive AI stock prices lower across the board but Citadel knew all along the Fed wouldn't actually hike.

Leopold Aschenbrenner's Situational Awareness fund, which had been up as much as 270% earlier in 2026 and grew past $20 billion in assets, got caught directly in that downdraft and the fund is reportedly now in crisis mode after its AI-related bets sank.

Citadel is now buying Situational Awareness's stock portfolio outright, essentially stepping in as the buyer of last resort for a fund that had built massive concentrated positions in names like CoreWeave, Core Scientific, Iris Energy, Vistra, and Applied Digital.

That portfolio was extremely top heavy, with CoreWeave alone making up over 25% of the fund's weight, Core Scientific over 16%, and Iris Energy over 15%, meaning when AI infrastructure names got hit in this rout, the concentration turned a sector pullback into a fund-threatening drawdown.

The dark irony people is that Citadel may have helped set the conditions for the selloff through its rate hike calls, then ended up buying the exact portfolio that got wrecked by it.

Congragulations Hedge funds win again and the only loser in all of this is retail who got manipulated and panic sold

Make sure to follow me @melvininvests for more insights!

See 8 related tweets

@MelvinInvests: If you haven't realized by now, Ken Griffin is a genius and the Situational Awareness deal is proof ...
@StockSavvyShay: Citadel reportedly bought most of Situational Awareness’s stock portfolio after the forced unwind.

...

@milesdeutscher: Insane story.

The TL;DR for those who don't know what happened:

Leopold Aschenbrenner (former Ope...

@ziv_ravid: A crazy storey. Leopold Aschenbrenner's fund went from $225M at launch in late 2024 to roughly$ 45B,...
@BullTheoryio: BREAKING: Situational Awareness just sold the bulk of its stock portfolio to Citadel.

The move come...

5. DynamicWebPaige (Group Score: 210.0 | Individual: 30.9)

Cluster: 9 tweets | Engagement: 34 (Avg: 130) | Type: Tech

📹 Video understanding with the robotics model is next level - actions, timestamps, and sensor readings cleanly annotated and tallied for any action in a clip.

Frontier-level performance, just at a fraction of the price (and much faster!).

You can track how many times a given action is performed, how many types of equipment enter an area, what sensors are within range and which aren't, and a whole bunch more:\n\nQT @GoogleAI: For decades, we’ve dreamed of robots that can seamlessly step into our world and lend a hand.

Today, we take a major stride toward making that dream a reality:

Introducing Gemini Robotics 2 from @GoogleDeepMind, the intelligence layer powering the next generation of truly adaptable robots. This major advance unlocks intelligent whole-body control, advanced dexterity, and even multi-robot collaboration 🤯.

Ok but... how does a robot actually "think"?

Real-world tasks take time and planning. To manage that complexity, our new embodied reasoning model, Gemini Robotics ER 2, acts as the robot’s high-level brain, enhancing the robot’s capabilities to:

— Observe the environment — Reason about the actions needed to complete the task — Coordinate with the vision-language-action model to carry out actions — Track progress until the job is done

This setup allows robots to execute complex multi-step workflows, self-correct if a step fails, and adapt to completely novel situations.

Learn more about Gemini Robotics ER 2 (and our two other brand new models) here: https://t.co/1YEpoYAhww

See 8 related tweets

@Scobleizer: RT @GoogleAI: For decades, we’ve dreamed of robots that can seamlessly step into our world and lend ...
@DynamicWebPaige: ⚡️ Frontier-level performance on robotics tasks, but faster and just a fraction of the cost!\n\nQT @...
@lukas_m_ziegler: Whole-body reasoning for humanoids! 🦿

@GoogleDeepMind released Gemini Robotics 2, three models work...

@Google: RT @OfficialLoganK: Introducing Gemini Robotics ER 2, our latest robotics embodied reasoning model b...
@VaibhavSisinty: Okay this is actually wild. Google gave two robots the same task. Nobody told them who does what. Th...

6. JamesMontemagno (Group Score: 192.0 | Individual: 49.9)

Cluster: 7 tweets | Engagement: 768 (Avg: 67) | Type: Tech

RT @satyanadella: Just wrapped our earnings call.

It was a very strong close to what was a record fiscal year for Microsoft.

· Annual revenue: $331B, +18% · MS Cloud:$ 214B, + 27% · And Azure: $100B, +41%

And even bigger opportunity ahead!

I wanted to share some more perspective on two areas of focus for us:

See 6 related tweets

@WOLF_Financial: MICROSOFT $MSFT JUST ANNOUNCED AZURE PASSED$ 100 BILLION IN ANNUAL REVENUE FOR THE FIRST TIME

CEO S...

@rohanpaul_ai: Microsoft just published FY26 Q4.

Profits increased 31%, robust cloud growth, Azure cloud sales m...

@StockMKTNewz: Microsoft's $MSFT Azure Cloud unit grew its revenue by 43% last quarter its fastest growth rate sinc...
@coinbureau: 🇺🇸TODAY: $MSFT surges 9% overnight after delivering a blockbuster quarter.

FY26 Q4 highlights:

R...
@Mayhem4Markets: $MSFT is surging after crushing earnings yesterday https://t.co/YnXvFyMWUO\n\nQT @Mayhem4Markets:$ M...

7. firstadopter (Group Score: 191.0 | Individual: 43.4)

Cluster: 6 tweets | Engagement: 562 (Avg: 179) | Type: Tech

WE ARE SO BACK! This sounds VERY GOOD for memory chip companies.

Samsung Electronics (h/t @jukan05 ): "Based on the demand visibility we currently have from customers, a substantial amount of unmet demand will roll over into next year, creating additional supply pressure. We expect the memory shortage in 2027 to be even MORE SEVERE than it is this year, with tight supply conditions likely to persist into 2028."\n\nQT @jukan05: From Samsung Electronics Q2 Earnings Call

Q: Do you expect the current memory shortage to persist into next year? If possible, could you also share your medium- to long-term outlook for memory demand?

A: The rapid acceleration of agentic AI is driving an explosive increase in token consumption. This is fueling unprecedented demand not only for AI servers but also for general-purpose computing servers.

In practice, AI frontier model developers that have been unable to secure sufficient cloud capacity from hyperscalers are now requesting allocations from neocloud providers as well. This has translated into large-scale memory procurement by server OEMs that primarily serve those neocloud customers.

Even so, memory shortages mean that many frontier AI companies are still unable to secure the infrastructure they need. To address this, they have begun sharing their medium- to long-term demand forecasts directly with us and expressing their intention to purchase memory from Samsung. We are also seeing the start of requests for long-term supply agreements (LTAs) to secure additional volume.

As the adoption of agentic AI continues to accelerate, memory demand is expanding at an exceptionally rapid pace. Industry supply remains well below demand. Even with increased industry-wide capex, it takes more than three and a half years from the construction of a new fab to wafer production. As a result, meaningful supply expansion through new capacity additions will take considerable time. We therefore believe a significant increase in industry supply before 2028 is unlikely.

Based on the demand visibility we currently have from customers, a substantial amount of unmet demand will roll over into next year, creating additional supply pressure. We expect the memory shortage in 2027 to be even more severe than it is this year, with tight supply conditions likely to persist into 2028.

Looking beyond 2029, it is still too early to make definitive projections. However, as AI token demand continues to surge, large customers building long-term AI infrastructure are expected to continue requesting multi-year supply agreements.

These long-term agreements are well aligned with our objective of hedging future business risks. We intend to prioritize contracts with customers that can provide firm, long-term demand commitments.

Over time, this should allow us to transition away from the historically cyclical nature of the memory industry toward a more stable and predictable business model.

With improved long-term demand visibility through LTAs, we will be in a better position to execute a more flexible supply strategy. Following our existing approach of securing cleanroom infrastructure in advance and installing production equipment in line with demand, we expect to further strengthen this disciplined and flexible capacity expansion strategy.

See 5 related tweets

@jukan05: From Samsung Electronics Q2 Earnings Call

Q: Do you expect the current memory shortage to persist i...

@jukan05: [Four Key Questions from Samsung Electro-Mechanics’ Conference Call]

Q1) (Company-wide) What is the...

@KyleReidhead: RT @KyleReidhead: It's VERY CLEAR that AI infra stocks will be MUCH HIGHER in the next few months

S...

@jukan05: Samsung Electronics Q2 Earnings Call Q&A

Q. What is the supply mix between HBM and conventional DRA...

@jukan05: Samsung Electronics 2Q26 Earnings Call Q&A

Q. What is the current foundry utilization rate?

A. Uti...

8. haider1 (Group Score: 188.9 | Individual: 27.3)

Cluster: 11 tweets | Engagement: 282 (Avg: 78) | Type: Tech

RSI is no longer sci-fi

GPT-5.6 Sol autonomously optimized its production kernels, cutting serving costs by 20%, while also improving its own draft model and boosting token efficiency by over 15%

this means openai is basically on track to reach its "AI intern" milestone by september\n\nQT @OpenAI: After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making itself more efficient to run.

The results:

20% lower serving costs from production GPU kernel improvements.
15%+ better token-generation efficiency from improved speculative decoding.

See 10 related tweets

@johncoogan: RT @OpenAI: After deployment, we applied GPT-5.6 Sol to advance the frontier of efficiency by making...
@ShanuMathew93: Are we now in an era of rapid revenue acceleration which everyone now knows and has seen given the...
@milesdeutscher: We're entering the token-efficiency era of AI.

GPT-5.6 Luna hits an intelligence score of ~51 at ro...

@simonw: GPT-5.6 found optimizations that "reduced end-to-end serving costs by 20%" for OpenAI to serve that ...
@dejavucoder: openai: we spent a billion dollah to train you. now that you have grown, its your time to give it ba...

9. wallstengine (Group Score: 178.5 | Individual: 32.8)

Cluster: 8 tweets | Engagement: 763 (Avg: 193) | Type: Tech

SAMSUNG Q2 OPERATING PROFIT JUMPS 19-FOLD TO A RECORD ON AI CHIP DEMAND

Samsung Electronics reported Q2 operating profit of 89.5T won, up from 4.68T won a year ago, while revenue rose 130% to 171.5T won.

Samsung expects strong H2 memory demand, led by continued AI infrastructure spending and broader agentic AI adoption, to keep the market undersupplied despite softer mobile and PC demand.

See 7 related tweets

@cryptopunk7213: just so i’m clear, ai semi stocks are down 30-50% across the board this month but:
samsung just r...
@CNBC: Samsung Electronics extended its record run with second-quarter operating profit topping analysts’ e...
@CNBC: Samsung Electronics second-quarter operating profit beats estimates on soaring AI chip demand https:...
@StockSavvyShay: SAMSUNG Q2 EARNINGS

• Revenue: $118.1B vs. Est.$ 119.2B (+130% YoY) • Operating profit: $61.7B vs. ...

@moneycontrolcom: #Business | Samsung chip profit jumps 250-fold on AI boom; costlier memory pushes mobile arm into fi...

10. vonderleyen (Group Score: 164.4 | Individual: 43.2)

Cluster: 5 tweets | Engagement: 14487 (Avg: 5880) | Type: Tech

AI is the most important technology of our time.

Europe wants to become the first AI Continent.

For advanced healthcare, for the transport sector and so much more.

European AI Gigafactories will provide the necessary computing power to make this possible.

Together with our Member States, we are funding their construction with up to €10 billion, which are set to unlock at least €20 billion in private investements across the EU.

Together, we are building our technological sovereignty.

https://t.co/Vxko4TRXVL

See 4 related tweets

@Altimor: That's almost 10% of all the money raised by OpenAI and Anthropic! Imagine all the environmental imp...
@TMTLongShort: https://t.co/3MhPiFsRAm\n\nQT @vonderleyen: AI is the most important technology of our time.

Europe...

@kimmonismus: „Europe wants to become the first AI Continent.“

They can't be serious, can they? This has to be a ...

@EU_Commission: We're bringing world-class computing power directly to European innovators, start-ups and researcher...

11. wallstengine (Group Score: 155.5 | Individual: 33.5)

Cluster: 7 tweets | Engagement: 110 (Avg: 193) | Type: Tech

GOOGLE BACKS ANTHROPIC’S TEXAS AI CAMPUS

Nexus Data Centers is in advanced talks for $15B in financing led by Morgan Stanley to build a 1.6GW campus and natural-gas power plant in Hubbard, Texas.

Anthropic has signed four data-center leases and corresponding power-purchase agreements. Google would guarantee billions of dollars in lease and power obligations if Anthropic defaults and is expected to receive roughly a 20% equity stake in the project.

The campus would use Google TPUs co-designed with Broadcom, with the chips financed separately through a vendor-financing agreement between Anthropic and Broadcom. The financing package includes a $14B bridge loan and a revolving credit facility.

Source: WSJ

See 6 related tweets

@MTSlive: SITUATION EXPLAINED: Banks are in talks to lend $15 billion for an Anthropic data center, with Googl...
@coinbureau: 🚨BREAKING: Banks are in talks to lend $15 BILLION to build an Anthropic data center in Texas, backed...
@StockMKTNewz: BANKS IN TALKS TO LEND $15 BILLION FOR ANTHROPIC DATA CENTER BACKED BY GOOGLE

A data-center develo...

@MeghanBobrowsky: RT @anissagardizy8: SCOOP: A data-center developer working with Anthropic is in advanced talks to bo...
@StockSavvyShay: $GOOGL is backing Anthropic’s planned 1.6GW Texas AI campus as Nexus seeks$ 15B to fund the data cen...

12. heyshrutimishra (Group Score: 154.9 | Individual: 46.6)

Cluster: 6 tweets | Engagement: 34100 (Avg: 897) | Type: Tech

RT @heyshrutimishra: WTF

AI companies are purchasing large quantities of used and rare books. Scanning their contents to train models. Then turning the originals to pulp.

The sum of human thought, digitized and shredded.

Every book that gets scanned disappears from the physical world permanently. The knowledge survives only as training data inside a system no one can hold, browse, or resell.

What used to sit on a shelf for centuries now exists as weights in a model that might get deprecated next quarter.

See 5 related tweets

@BrianRoemmele: RT @BrianRoemmele: In this machine I can scan 1000s pages a day and never destroy a book.

No guillo...

@pitdesi: Most people mad about this seem to misunderstand it

Rare doesn’t mean collector’s item. It means ou...

@alex_prompter: Project Panama by @AnthropicAI is real https://t.co/9w086O6ZlA\n\nQT @heyshrutimishra: WTF

AI comp...

@ProudSocialist: AI companies are buying up books in bulk, scanning them to train their AI models, and then destroyin...
@venturetwins: One of the strangest misinformation campaigns around AI right now is that the labs are buying physic...

13. UnslothAI (Group Score: 151.2 | Individual: 31.1)

Cluster: 7 tweets | Engagement: 580 (Avg: 1330) | Type: Tech

You can now run Inkling-Small, a new 276B model by Thinking Machines.

Inkling-Small is the strongest open model for its size and runs local on 128GB RAM.

Apache-2.0 Licensed, it has image, audio + 1M context support.

Guide: https://t.co/mwyyMtrtA9 GGUF: https://t.co/Mrh0giFSTH https://t.co/p8ChQ3Hp95\n\nQT @thinkymachines: Today, we are releasing Inkling-Small.

Inkling-Small achieves comparable performance to Inkling at a quarter of its size. It features 276B total parameters, 12B active. We are making the full weights available.

https://t.co/BtYNcpkDRA

Fine-tune it on Tinker today, or chat with it in text, image, and audio on Tinker Playground.

See 6 related tweets

@Designarena: Inkling Small by @thinkymachines is now available on Design Arena!

Built as a Mixture-of-Experts mo...

@simonguozirui: Inkling-Small? more like Shrinkling 👀\n\nQT @thinkymachines: Today, we are releasing Inkling-Small. ...
@ziv_ravid: Thinking machines releasing Inkling-Small. 276B total parameters with only 12B active (a quarter of ...
@0xSero: Let’s go perfect for 2-4 sparks / 2-4 RTX Pro 6000 / 8x 3090/4090

Same footprint as Deepseek-v4-Fla...

@mervenoyann: Thinking Machines released Inkling Small (🦖) + NVFP4

12B active 276B total params, the model perfor...

14. ANI (Group Score: 142.5 | Individual: 52.0)

Cluster: 4 tweets | Engagement: 1247 (Avg: 224) | Type: Tech

#WATCH | Bengaluru, Karnataka: Co-founder, Sarvam AI, Pratyush Kumar says, "... It's been three years now. We started off building on top of the Llama model, which was the best open model at that time. Sarvam was the only company outside, or maybe in Asia, that contributed to the Llama 3 Series. We understood how to add language skills into models, and Llama 3 was reasonable in Indian languages. When the first set of GPUs arrived in India, we pre-trained from scratch a 2-billion model and worked with Infosys to make it useful for IT services. Then we recognised the importance of scaling up reinforcement learning and built a reasoning model called Sarvam M on top of Mistral's small model. Over the last year, we pre-trained from scratch in India on 4,000 Hopper Series GPUs a 100-billion-plus-parameter model..."\n\nQT @ANI: #WATCH | Bengaluru, Karnataka: Bengaluru-based startup Sarvam AI launched Epoch Builder Edition, a platform that enables developers and enterprises to build large language models for Indian languages and supports more than 10 Indian languages.

Sarvam AI focuses on building generative AI models and solutions tailored for Indian languages and diverse local use cases.

See 3 related tweets

@ANI: #WATCH | Bengaluru, Karnataka: Bengaluru-based startup Sarvam AI launched Epoch Builder Edition, a p...
@ANI: #WATCH | Bengaluru, Karnataka: Co-founder, Sarvam AI, Pratyush Kumar says, "... Let me begin with wh...
@beatsinbrief: 🚨 BREAKING: Sarvam AI has launched Epoch Builder Edition, a new platform that helps developers and e...

15. financialjuice (Group Score: 141.4 | Individual: 26.8)

Cluster: 6 tweets | Engagement: 78 (Avg: 38) | Type: Tech

🔴$AMZN Amazon Q2 Earnings & Q3 Guidance

EPS $5.75 est.$ 1.99 Net Sales $200.6B, est.$ 197.01B Sees Q3 Net Sales $197.0B–$ 202.0B, est. $203.93B Sees Q3 Operating Income$ 22.5B– $26.5B, est.$ 25.07B Operating Income $27.46B, est.$ 23.61B Operating Margin 13.7%, est. 12.0% AWS Net Sales $42.23B, est.$ 40.57B AWS Net Sales (ex-FX) +37%, est. +31.3% North America Net Sales $116.18B, est.$ 113.94B Physical Stores Net Sales $5.79B, est.$ 5.87B AWS AI business exceeded a $25B annual revenue run rate Trailing 12-month free cash flow declined to an outflow of$ 7.6B AI & Chips businesses each exceeded annual revenue run rates of over $25B Q3 guidance includes an unfavorable FX impact of 80 bps

See 5 related tweets

@MilkRoadAI: RT @MelvinInvests: Amazon just dropped a blowout quarter and AWS is the reason why (Save this).

Ama...

@WOLF_Financial: AMAZON $AMZN JUST REPORTED Q2 EARNINGS

• Revenue: $200.6B, beating expectations of$ 196.47B 🟢 • EPS...

@WOLF_Financial: AWS IS 21% OF AMAZON'S REVENUE AND MORE THAN 60% OF ITS OPERATING PROFIT

Everyone still calls Amazo...

@StockMKTNewz: Amazon $AMZN just said its AI business for Amazon Web Services has exceeded a$ 25 Billion annual rev...
@StockMKTNewz: Amazon Web Services $AMZN is now a$ 168.8 Billion Revenue Run Rate business

AWS grew by 36.8% duri...

16. itsolelehmann (Group Score: 136.4 | Individual: 37.0)

Cluster: 4 tweets | Engagement: 2894 (Avg: 2347) | Type: Tech

should be obvious by now, but OpenAI and Anthropic are just gonna keep cannibalizing all their biggest customers.

it’s simply too profitable for them to resist. and it’s already happening:

Figma partnered with Anthropic on AI design tools. then Anthropic’s product chief quit Figma’s board, and 3 days later Anthropic launched Claude Design to compete with Figma. CEO Dylan Field said Anthropic was “not consistently candid.”
Novo Nordisk uses Claude to help develop drugs. now Anthropic is developing drugs of its own.
Microsoft poured billions into OpenAI. now OpenAI is building a Jobs Platform to compete with LinkedIn, and reportedly a code repository to compete with GitHub. Microsoft owns both.
Harvey uses Claude to sell AI contract analysis, due diligence, and litigation tools. now Anthropic sells those same workflows through Claude for Legal.
Intercom used OpenAI’s Realtime API to build Fin Voice. now OpenAI sells its own voice-and-chat support agent through Presence.
Abridge and Ambience build clinical documentation products on OpenAI. now OpenAI sells ChatGPT for Healthcare directly to hospitals with clinical documentation built in.
Benchling uses Claude to power its biotech R&D platform. now Anthropic sells its own scientific workbench through Claude Science.

the frontier lab playbook is simple:

sell their models to the world’s most valuable businesses
help wire them into those companies’ most valuable and sensitive work
map the business from the inside and find where AI can take over
turn those capabilities into their own products and become the customer’s competitor\n\nQT @chamath: It’s well within Anthropic’s rights to compete in any market they choose.

What’s funny, in this instance, are the number of Pharma companies, who through their unchecked use of Anthropic, are driving revenues into what they think is a model provider but is in fact a competitor lurking in the shadows thereby accelerating their own demise.

I suspect any end market with reasonable ROCE that could be AI accelerated is on the table.

If I were them, I’d probably do the same.

See 3 related tweets

@HarryStebbings: "The old idea of 'negative signalling' in venture capital has largely broken down.

With larger roun...

@mervenoyann: what would've happened if these companies adopted open weights from the get go 🌝\n\nQT @itsolelehman...
@MikeBradleyAI: RT @itsolelehmann: should be obvious by now, but OpenAI and Anthropic are just gonna keep cannibaliz...

17. BanghuaZ (Group Score: 134.7 | Individual: 30.1)

Cluster: 6 tweets | Engagement: 43 (Avg: 22) | Type: Tech

Excited to work with Google cloud @googlecloud on bringing SGLang to TPU!\n\nQT @radixark: RadixArk and Google Cloud are joining forces with the SGLang community to make TPU a drop-in, cost-efficient path to frontier inference.

SGL-JAX already serves the major open model families on the latest TPU generations: Gemma, Qwen, DeepSeek, GLM, Kimi, Ling, MiniMax, MiMo, Grok, plus Wan and Flux for video and image generation.

With this partnership, TPU unlocks SGLang's production feature set: 5D parallelism, Radix Cache, HiCache, quantization, speculative decoding, on TPU Pallas kernels co-built by Google, RadixArk, and the SGLang community. SGL-torchtpu opens a PyTorch-native path to all of it.

SGLang on TPU comes with the same API and the same features, so for teams already running it, getting started is simple. Going forward, TPU support for new open models will land alongside the rest, and that carries forward to each new TPU generation.

Full announcement and SGL-JAX repo 👇

See 5 related tweets

@ying11231: RadixArk's mission is to make frontier AI infrastructure open and accessible to every builder. SGLan...
@richardczl: Step by step, keep growing!\n\nQT @radixark: RadixArk and Google Cloud are joining forces with the S...
@lmsysorg: RT @radixark: RadixArk and Google Cloud are joining forces with the SGLang community to make TPU a d...
@googledevs: Big news: @Google and @RadixArk are partnering to bring @sgl_project to Google Cloud TPUs!

✅ Run S...

@ying11231: RT @sgl_project: Excited to see RadixArk and Google joining the SGLang community to work on TPUs!
...

18. WhaleInsider (Group Score: 132.6 | Individual: 41.0)

Cluster: 4 tweets | Engagement: 4652 (Avg: 602) | Type: Tech

JUST IN: 🇮🇳 India’s Sarvam announces it’s developing a trillion-plus parameter AI model from scratch that’s five times cheaper than global rivals. https://t.co/Jh3bNjCiAH

See 3 related tweets

@moneycontrolcom: #AIWithMC | Sarvam to build one trillion parameter model, says its pricing is 5.5 times cheaper than...
@chandrarsrikant: Follow @BhavyaDKumar for all the updates from Sarvam Epoch ⏬ @SarvamAI Co founder Pratyush announce...
@_TheTathya: 🚨 Sarvam AI Unveils India’s Trillion-Parameter Answer to Global AI Giants

India’s sovereign AI push...

19. teortaxesTex (Group Score: 132.5 | Individual: 34.4)

Cluster: 4 tweets | Engagement: 43 (Avg: 64) | Type: Tech

Andrew flew too close to the sun… And is going even further beyond Very important line of work\n\nQT @andrewho03: Today is my last day at @OpenAI. I'm glad to have spent the last eight months of my life working here!

I'm starting a new company focused on the production of high-quality reinforcement learning datasets:

The generalization ability of LLMs is clearly very poor, with "spiky" capabilities even in areas that have received tremendous amounts of investment and attention. For example, despite multiple years with tens (if not hundreds) of billions invested, even coding capabilities don't demonstrate "generality" -- even if every model can solve Codeforces questions or port C++ to Rust better than I can, I still have to manually "deslop" pull requests.
The vast majority of economically productive capabilities are not well represented in existing data offerings. First, there's a certain art to the design of an RL dataset which most vendors, not having upstreamed data into large training runs themselves, don't really understand. Second, and more importantly, most work is highly contextual and not easily encoded into a gradable environment; even if we can observe a "golden path" taken by a human which we believe to be good, it's challenging to understand whether alternate, counterfactual paths produce good or bad outcomes.

The basic premise here is that I have a clear understanding of what labs need/want, having explicitly been on the other side and having been involved at every level from procurement all the way through training, and I'm able to provide it. I also believe that data needs will grow tremendously in the coming years, especially as frontier labs face increasing pressure toward profitability, and that they won't get the relevant capabilities "for free" through scaling alone; instead, they'll need to spend >$100B on precise, well-targeted data acquisition.

Our first products will be focused on biology and statistical reasoning:

First, datasets that address long-horizon scientific reasoning, drawing on my work on GeneBench-Pro with @jeremyli__. Frontier models are still unable to reliably execute "messy" data analyses that require judgment, exploration, and adaptive revision (GB-Pro passrate on GPT-5.6 Sol scarcely exceeds 30%); to address this, we have the ability to generate thousands of high-quality problems with known ground truths which can be reliably graded. (In contrast, most existing RL data for bioinformatics is either massively over- or under-specified, and will probably break your model when you train on it.) Moving the "reliability gap" from 30% to >90% is obviously required for scientific acceleration, and -- despite my skepticism about generalization of RL -- is one of the most promising datasets conceivable when it comes to yielding generalization benefits for models' overall reasoning capabilities.
Second, datasets that address capabilities relevant to day-to-day workflows. Imagine a scientist snapping a picture of some experimental process or result -- say, a cell culture plate or a Western blot -- and asking Claude a question. Frontier models remain quite bad at these questions, especially those with multimodal components. But they're obviously required for acceleration of scientific discovery; before we can dream about automating science, we have to begin with shoring up these basic, generalist capabilities.

Beyond these two, we hope to expand to adjacent fields (chemistry, materials science, etc.), and then even further into fields with more direct economic applicability like healthcare and white-collar office work.

I strongly encourage labs with data needs to reach out. We offer industry-standard pricing and terms, and like I said -- I know how this process works, what good data looks like, and how to demonstrate to you, convincingly, that you'll be able to upstream our data into your training processes without issue. My DMs are open!

See 3 related tweets

@jon_stokes: RT @andrewho03: Today is my last day at @OpenAI. I'm glad to have spent the last eight months of my ...
@andersonbcdefg: interesting\n\nQT @andrewho03: Today is my last day at @OpenAI. I'm glad to have spent the last eigh...
@alexframegreen: This is very well written. I would invest. Very clear that there is a strong incentive to stay on t...

20. WesRoth (Group Score: 131.8 | Individual: 36.0)

Cluster: 5 tweets | Engagement: 16 (Avg: 32) | Type: Tech

Thinking Machines cofounder Lilian Weng is returning to OpenAI.

Weng stepped down from Mira Murati’s startup after saying the pace of a cofounder role had become unsustainable for her health.

She will now lead an OpenAI team focused on accelerating the company’s internal research using AI—part of its effort to develop systems capable of helping train and improve future models.

Weng previously worked at OpenAI from 2018 to 2024.

She contributed to GPT-4 across pretraining data, evaluations, safety, and deployment, built the Applied AI Research team, and later led Safety Systems.

Her departure leaves only Mira Murati and John Schulman from Thinking Machines’ original six cofounders.

Barret Zoph and Luke Metz previously returned to OpenAI, while Andrew Tulloch joined Meta.\n\nQT @steph_palazzolo: Lilian Weng, the Thinking Machines cofounder who announced her departure earlier this week, citing startup-related stress and illness, is rejoining OpenAI.

She'll be working on using AI to develop new models (recursive self-improvement).

w/ @amir

https://t.co/rart1w6KkM

See 4 related tweets

@rohanpaul_ai: Techcrunch: Lilian Weng is leaving Thinking Machines Lab for OpenAI, moving from co-founder role int...
@Michaelzsguo: Per The Information, less than two days after announcing her departure, Lilian Weng has already fina...
@teortaxesTex: Lilian caught a debilitating strain of FOMO Vulgaris\n\nQT @steph_palazzolo: Lilian Weng, the Thinki...
@rohanpaul_ai: RT @rohanpaul_ai: Techcrunch: Lilian Weng is leaving Thinking Machines Lab for OpenAI, moving from c...