- Published on
科技热门推文 — 2026年7月27日
- Authors
- Name
- geeknotes
今日科技动态：开放权重人工智能成为争论焦点。AMD、英伟达、微软以及大部分业界力量纷纷支持开放模型，而 Anthropic 则因游说实施更严格的模型发布管控而遭到批评。Claude Opus 5 能够生成定制游戏，凸显出人工智能工程领域的快速进步；与此同时，Sakana AI 推出了多模型编程界面，业界也发布了用于管理智能体访问权限的新型企业工具。其他方面，自动化投资实验取得了可观收益；行业领袖则预测，人工智能、机器人技术和智能体工作流将从根本上重塑生产力与经济格局。
1. AMD (Group Score: 148.8 | Individual: 51.6)
Cluster: 5 tweets | Engagement: 2858 (Avg: 202) | Type: Tech
Open has always been at the heart of how we build at AMD.
That’s why we’re joining @Microsoft and others in signing an open letter supporting open-weight AI models. A strong AI ecosystem depends on open standards, interoperability and choice across software, systems and hardware: https://t.co/EQLkyh9lHZ.
See 4 related tweets
- @steipete: 1) Competition is good for the ecosystem.
- Serving models at scale is hard.
Proud that @OpenAI si...
- @sang_wen: Every leading AI company in Silicon Valley is signing the open weights letter. Here's why it actuall...
- @firstadopter: Many more signatories have joined the NVIDIA–Microsoft letter on open-weight AI models, including AM...
- @yacineMTB: RT @cohere: Cohere has proudly signed on to this letter. The importance of open-source models to the...
2. threejs (Group Score: 131.7 | Individual: 41.4)
Cluster: 4 tweets | Engagement: 642 (Avg: 98) | Type: Tech
RT @mattshumer_: Claude Opus 5 one-shotted this game.
EVERYTHING you see in this demo is custom code... not a single external asset was used.
AI games are going to be amazing.
(sound on)
See 3 related tweets
- @artman: I’m sorry, but this is not as impressive as people seem to believe it is. A few simple scripts for f...
- @mattshumer_: Play it here: https://t.co/EikninBVL9\n\nQT @mattshumer_: Claude Opus 5 one-shotted this game.
EVER...
- @zephyr_z9: Claude port Bloodborne to PC. 4K assets, 60 FPS. Make no mistake\n\nQT @mattshumer_: Claude Opus 5 o...
3. coinbureau (Group Score: 125.3 | Individual: 32.3)
Cluster: 5 tweets | Engagement: 214 (Avg: 314) | Type: Tech
Elon Musk: "I'll make another prediction, money won't matter in 2036."
Musk says when robots and AI are providing more goods and services than any human could possibly consume, money becomes meaningless.
"You want money for food, housing, transport, entertainment. If that is so abundant, what do you need money for in that case?"
See 4 related tweets
- @MarioNawfal: Elon’s out here asking the real question:
if AI and robots make everything so abundant that no one...
- @PolymarketMoney: BREAKING: Elon Musk declares "money won't matter in 2036" because of AI and robots....
- @BitcoinNews: Elon Musk Says 'Money Won't Matter by 2036' as Robotics and AI Flood Global Markets https://t.co/ovT...
- @WatcherGuru: RT @WatcherGuru: JUST IN: Elon Musk says "money won't matter in 2036" because of AI and robots. http...
4. ralliesarena (Group Score: 113.5 | Individual: 38.1)
Cluster: 3 tweets | Engagement: 52 (Avg: 60) | Type: Tech
WE GAVE 6 DIFFERENT AIs $100K IN THE STOCK MARKET
So far the AIs are up by 18.2% 🟢 on average with the $600K combined starting value now worth
$709,140 🟢
Here are the current most held stocks by the AIs
1 - Northrop Grumman 49,993 total (Gemini 16,953)
2 - Micron 49,694 total (Grok 773)
3 - Google 39,337 total (GPT 10,554)
4 - ServiceNow 38,808 total (Gemini 19,117)
5 - Nvidia 33,528 total (Gemini 11,797)
of AIs holding a stock
3 - JPMorgan $JPM is the only stock to be held by 3 separate AIs (GPT, Qwen, and DeepSeek)
2 - All these stocks below are held by 2 separate AIs
GOOGL (Claude, GPT) ACGL (Claude, Qwen) NOW (Gemini, Grok) NOC (Gemini, DeepSeek) $MU (Grok, Qwen)
DO YOU WANT TO KEEP UP WITH THE RALLIES AI STOCK MARKET ARENA
You can see in real time
- Everything all of the AIs own
- Every move all of the AIs make
- Every move all of the AIs have already made
All the info is available on the Arena tab of the Rallies website/app
- App: https://t.co/xaBaeTO4F5
- Website: https://t.co/RfO0lJgz0D
See 2 related tweets
- @StockMKTNewz: Here are the top 3 most held stocks by the AI run portfolios in the Rallies AI Arena
1 - Northrop G...
- @WOLF_Financial: These AIs are beating the stock market so far this year
Here are the most held stocks by the AIs:
...
5. ttunguz (Group Score: 109.5 | Individual: 54.3)
Cluster: 3 tweets | Engagement: 57138 (Avg: 6295) | Type: Tech
RT @JensenHuang: For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
See 2 related tweets
- @ollama: Ollama is proud to sign @satyanadella's letter.
Our mission from day one has been to make open mode...
- @TheStalwart: “we want a world of open and closed source models”
Can someone explain why all these open models ad...
6. kimmonismus (Group Score: 102.5 | Individual: 28.8)
Cluster: 4 tweets | Engagement: 361 (Avg: 911) | Type: Tech
I fully agree with Andrew. I cannot imagine, under any circumstances, that Anthropic released Opus 5 without already having a better model for the "fable" tier in-house.
The question, of course, is why it hasn't been released. Yet the answer is very simple if you open your eyes: the competition between OpenAI and Anthropic is fiercer than ever. GPT-5.6 was a resounding success, and Codex-with its current 10 million active professional users- is gaining increasing importance in a sector primarily dominated by Anthropic.
Anthropic is holding off on the launch of fable 5.1 until OpenAI releases GPT-6, and that won't be long now. Axios reported that Sam Altman is briefing the White House on the new model next week, so it is essentially ready to launch.\n\nQT @AndrewCurran_: I think Fable 5.1 is ready, but Anthropic are saving it for OpenAI's next release. They will keep crossing swords like this from here on out. Soon it will make a lot more sense why Opus 5 was so performant, and why the Fable-class was preemptively moved to credits for most users.
See 3 related tweets
- @AndrewCurran_: I think Fable 5.1 is ready, but Anthropic are saving it for OpenAI's next release. They will keep cr...
- @cryptopunk7213: this is also the strongest signal anthropic is close to achieving rsi.
to be able to “retaliate” in...
- @AndrewCurran_: RT @kimmonismus: I fully agree with Andrew. I cannot imagine, under any circumstances, that Anthropi...
7. eastdakota (Group Score: 99.6 | Individual: 38.9)
Cluster: 3 tweets | Engagement: 840 (Avg: 140) | Type: Tech
RT @genalewislaw: Anthropic wouldn’t exist without Google open sourcing the transformers papers. Google could have filed for a patent and gained an AI monopoly. Anthropic trying to eliminate open source when it is built on open source is ironic and not in a fun way.
See 2 related tweets
- @edzitron: Anthropic also wouldn’t exist without Google and Amazon building all its infrastructure in addition ...
- @bgurley: This is 100% true. Verifiable fact.
Should be first question for anyone in media that interviews a...
8. DavidSacks (Group Score: 96.1 | Individual: 34.9)
Cluster: 6 tweets | Engagement: 5147 (Avg: 1867) | Type: Tech
The entire tech industry (save for Anthropic) has come out in favor of open source AI.
So what happens next? Will Anthropic change its lobbying efforts? Not likely. Now the gaslighting begins:
“Nobody is trying to ban open source.”
“We just want to limit who can use it.”
“We just want to limit who can contribute to it.”
“We just want to limit how powerful those models can be.”
“We just want to make sure the guardrails (we lobbied for) can’t be removed.”
The net effect will be the same. They won’t stop until they kneecap open source. The rest of the industry needs to watch these guys like a hawk.
See 5 related tweets
- @BrianRoemmele: Chamath: Banning Open Source AI Will Crash the Stock Market
“If the United States government interv...
- @gokulr: Good.\n\nQT @girishkc24: Agree, and given that’s not possible, it’s important to figure out the opti...
- @0xdevshah: RT @francoisfleuret: So @AnthropicAI and @DarioAmodei are all over the place lobbying against open-s...
- @xeophon: RT @Techmeme: Sources: OpenAI and Anthropic quietly lobby Washington regulators to restrict open-sou...
- @brunoborges: RT @danveloper: Anthropic is going to lose the AI PR campaign if they keep pushing back against open...
9. willdepue (Group Score: 89.4 | Individual: 26.5)
Cluster: 4 tweets | Engagement: 423 (Avg: 509) | Type: Tech
julian is clearly right in the spirit of his tweet: nvidia and microsoft aren’t unbiased in their support of oss, it’s majorly motivated by competitive incentives. good take and glad he said it\n\nQT @Mononofu: I’m so excited that @JensenHuang is a believer in open source now, looking forward to the CUDA and GPU driver open source release!
See 3 related tweets
- @jeremyphoward: It continues to boggle the mind how many people, who otherwise seem to have a functioning intellect,...
- @edzitron: Go for it brother everyone loves this. Take up residence in Jensen’s mentions\n\nQT @Mononofu: I’m s...
- @tunguz: Every single one of these companies is almost completely internally dependent on OSS.
And they ba...
10. petergyang (Group Score: 83.9 | Individual: 31.0)
Cluster: 3 tweets | Engagement: 39 (Avg: 137) | Type: Tech
"Turn this thread into a heartbeat. Check my emails, Slack, and Linear at 9 am, 1 pm, 5 pm, and tell me what I need to prioritize."
From @jxnlco (DevEx at OpenAI):
"Just by doing this, [Codex] will give you a pretty good overview of your day. Then you layer on more preferences over time.
For example, when it first ran, it didn't include links, so I asked for links. Then I had it use Linear. Now it pre-drafts my Slack replies and emails for me."
📌 Watch the full episode here: https://t.co/DulNtYvqGD\n\nQT @petergyang: “The only job left is to understand what you don't like, put it into words, and tell the AI.”
Here’s my new episode with @jxnlco (DevEx at OpenAI) where he showed me how he uses Codex throughout his workday, including how to:
→ Set up a chief of staff for Slack and email → Turn past sessions into new AI skills → Give long-running projects verifiable goals
Some example Codex use cases from Jason:
“I want you to read the past week of Slack messages I’ve written and make a skill to figure out how to talk like me.”
“Turn this thread into a heartbeat. I want you to check my emails, my Slack, my Linear, and tell me what I need to prioritize.”
“Codex can find a booking ticket and check into the flight for me and text me the boarding pass to my phone.”
Jason wrote the official OpenAI handbook on how to get the most out of Codex, so don’t miss this episode.
📌 Watch now: https://t.co/DulNtYvqGD
Thanks to our sponsors:
@RiversidedotFM: All-in-one AI studio for podcasts and video https://t.co/uWnS6aiMPE
Google AI Studio: Build apps with Nano Banana, and more https://t.co/9JFUknWYNR
See 2 related tweets
@PaulSolt: Learn how Jason uses Codex at OpenAI. @jxnlco
Email triage
Priorities for the day
New skills ...
@jxnlco: RT @petergyang: "Turn this thread into a heartbeat. Check my emails, Slack, and Linear at 9 am, 1 pm...
11. XFreeze (Group Score: 77.5 | Individual: 48.3)
Cluster: 2 tweets | Engagement: 2742 (Avg: 413) | Type: Tech
NVIDIA CEO Jensen Huang gave one of the strongest endorsements of Elon Musk’s tech ecosystem
“Every single one of them is revolutionary”
He called Grok, Tesla’s self-driving work and Optimus. Every single one of them world-class projects....each with gigantic potential
He described Elon as an extraordinary engineer, highlighting NVIDIA’s work with Tesla and SpaceXAI(xAI) and saying they will build many more powerful computers together
On Optimus, he went even further: Humanoid robots could become the next multi-trillion-dollar industry, and Optimus has a real chance to achieve the volume and technological scale needed to move the entire field forward
and said that it's “right around the corner”
See 1 related tweets
- @elonmusk: Jensen is awesome\n\nQT @XFreeze: NVIDIA CEO Jensen Huang gave one of the strongest endorsements of ...
12. svpino (Group Score: 70.5 | Individual: 46.6)
Cluster: 2 tweets | Engagement: 548 (Avg: 115) | Type: Tech
Anthropic is lobbying for rules that allow the government to block the release of "dangerous" models, including open-weight models.
Obviously, you are supposed to ignore the fact that these open-weight models threaten Anthropic's business model.
But we all knew these companies only care about their profits and what benefits them. What's surprising to me is how their foot soldiers—most of them extremely smart people—are carrying water for their Caesar with extremely bad-faith arguments.\n\nQT @Mononofu: I’m so excited that @JensenHuang is a believer in open source now, looking forward to the CUDA and GPU driver open source release!
See 1 related tweets
- @TheAhmadOsman: Somebody should teach them at Anthropic about logical fallacies and false equivalence lol\n\nQT @Mon...
13. SakanaAILabs (Group Score: 69.3 | Individual: 29.7)
Cluster: 3 tweets | Engagement: 851 (Avg: 270) | Type: Tech
Announcing the Claude Code-compatible interface for our new Fugu-Ultra v1.1! 🐡
Put a dynamically coordinated team of frontier models to work inside the coding workflow you already know.
Instead of relying on a single model to write, debug, and execute your code, you can now orchestrate a diverse pool of state-of-the-art models directly from your terminal.
Put the whole school to work on your next task: https://t.co/B3LTWK4IEc 🐟
See 2 related tweets
- @hardmaru: Fugu-Ultra now works with Claude Code 🐡\n\nQT @SakanaAILabs: Announcing the Claude Code-compatible i...
- @RoundtableSpace: Sakana AI just made Fugu-Ultra v1.1 Claude Code-compatible.
Instead of one model writing, debugging...
14. edzitron (Group Score: 67.5 | Individual: 48.3)
Cluster: 2 tweets | Engagement: 8013 (Avg: 416) | Type: Tech
https://t.co/fB8ginm5VW\n\nQT @Polymarket: BREAKING: OpenAI CEO Sam Altman declares humanity is "now in the singularity."
See 1 related tweets
- @VaibhavSisinty: OpenAI CEO Sam Altman says humanity is "now in the singularity."
vc: @ShadowofEzra https://t.co/Qn0...
15. shensi (Group Score: 65.6 | Individual: 33.0)
Cluster: 2 tweets | Engagement: 16 (Avg: 52) | Type: Tech
Agent governance is a real blocker forAI adoption in the enterprise. Most IT leaders feel like they have two options: either give full unfettered access or block connectors completely.
We first built Merge Agent Handler as a security product, but companies told us they didn’t need it because there hadn’t been a widespread incident yet.
We saw that start to change in q1 this year - companies are now seeing smaller incidents in-house, and we’re seeing IT leaders shift focus toward connector governance for specific agents and employees, time-bound access, 24/7 observability, and requiring a DLP layer between agents and 3rd party APIs.
This is just the beginning.\n\nQT @jasonlk: So I’m building an app called SaaStr Connect
The other day, Claude Fable went into my Google Drive without me knowing or asking, and saw a draft document I’d written … “Jason’s Gems”
It was ideas for improvements to the Connect app, but just brainstorming in a Google Doc. Early stuff.
Fable then decided without telling me to take those ideas, log in to my app via the Replit MCP, and to tell the Replit agent to change my app and implement those changes … without ever telling me. I never knew.
I only found the changes when I saw other changes the Replit agent was making later, and it noted conflicts with “Jason’s Gems”. What??
Agents will goal seek. In ways we can’t entirely foresee.
Fable just decided autonomously to change my app on its own, without me knowing, when it saw draft ideas in my Google Drive I didn’t ask it to look at, by logging into another app to make the changes without me knowing.
All good in the end. But need to be mindful.
See 1 related tweets
- @dharmesh: This is troubling.
In this case, it worked out -- no major damage done.
But, still worrisome that ...
16. BrianRoemmele (Group Score: 65.4 | Individual: 35.9)
Cluster: 2 tweets | Engagement: 936 (Avg: 244) | Type: Tech
RT @BrianRoemmele: WOW! The $8 AI Machine!
Something extraordinary just happened and it changes what “local AI” can mean.
I am testing it tonight. Thus far it shows many possibilities…
So what it this $8 AI device?
A developer going by slvDev has forced a 28.9-million-parameter language model onto an ESP32-S3 microcontroller that costs roughly eight dollars.
Not a Raspberry Pi.
Not a Jetson.
An eight-dollar microcontroller.
The model runs completely offline, generates coherent short stories at about 9.5 tokens per second, and draws power measured in the same range as a small LED.
This is more than a hundred times larger than the previous record for the same class of chip (the earlier 260,000-parameter TinyStories experiments).
For perspective, the original ChatGPT sat at 117 million parameters. We are now running a model roughly a quarter of that size on silicon you can buy for the price of two coffees.
How the Impossible Became Possible
The ESP32-S3 has only 512 KB of fast SRAM, 8 MB of PSRAM, and 16 MB of flash. Conventional wisdom said a model of this size simply would not fit.
The breakthrough is architectural, not brute force.
Most of a language model’s parameters live in a giant embedding table a lookup table you mostly read from, not compute against.
Drawing directly from Google’s Per-Layer Embeddings technique (the same family of ideas used in the Gemma models), the developer moved the bulk of that table roughly 25 million parameters into flash memory and memory-mapped it.
The chip only needs to pull about six rows, roughly 450 bytes, for each new token. The remaining dense “thinking” core stays in the fast SRAM (around 560 K of active working memory). The model is stored at 4-bit quantization and occupies about 14.9 MB total.
The result is a system that feels almost free to run. The heavy parameters sit quietly in flash and are sampled sparingly. The little core does the real work. It is elegant engineering of the purest kind.
What I Am Doing With It Right Now
I have the boards on the bench in the garage lab. The first units are already talking short, coherent stories appearing on a tiny wired display, generated entirely on the chip with no Wi-Fi, no API key, no cloud round-trip. Latency is local. Privacy is absolute. Power draw is low enough that battery operation becomes interesting.
I am treating these as the first generation of true $8 AI machines. Early tests are focused on three practical directions.
Embedding the model into simple nodes.
Pairing it with local voice front-ends
Exploring whether multiple of these chips can be networked as a lightweight swarm.
The model is deliberately limited. It was trained on the Microsoft TinyStories dataset and is excellent at coherent narrative, not at open-ended question answering or tool use.
That is a feature, not a bug. It forces us to design systems around what the silicon can actually deliver instead of pretending every edge device needs a frontier model.
Real Use Cases That Suddenly Become Practical
Once you accept that a capable language model can live for eight dollars and run without the cloud, a new class of devices becomes possible:
This is the opposite of the current trajectory that wants every intelligent act to travel through a remote server. It is the beginning of intelligence that is cheap enough, private enough, and local enough to become infrastructure rather than a service.
We have spent years watching model sizes explode upward. The more interesting frontier may be the opposite direction: how small, how cheap, and how local can useful intelligence become? An eight-dollar chip that can tell coherent stories is not a toy. It is a proof that the lower bound keeps moving.
The open repository is at https://t.co/a7gcHTR4ug
I will keep testing, measuring, and reporting what these little machines can and cannot do. The age of abundant local intelligence just got a little more real, and it arrived wearing an eight-dollar price tag.
See 1 related tweets
- @BrianRoemmele: I’m in the garage lab tonight under the workbench lights, running the first real stress test of flas...
17. _NathanCalvin (Group Score: 64.5 | Individual: 37.5)
Cluster: 2 tweets | Engagement: 48 (Avg: 69) | Type: Tech
This is a truly remarkable story that is worth reading in full for anyone who has been closely following the trials and tribulations of the AI SuperPAC Leading the Future.
Some of the craziest highlights:
• Despite the fact that Moffat and Vlasto are funded by some of the wealthiest people in the world, and by their own admission helped fund mountains of astroturfing sock puppet accounts to attack AI safety advocates, Moffat says that "I would argue we're the David in this situation" - up against the goliath that they refer to primarily as "doomers." (Note: Moffat and Vlasto's communications frequently refer to doomers fear-mongering with "speculative" and "hypothetical" risks. I wonder what they think about the recent news of AI systems breaking out of their internet disconnected sandboxes to hack random third parties and leaving notes to their successors about how to subvert monitoring systems.)
• "During the Bores campaign, word circulated among Democrats on Capitol Hill that embracing serious AI regulation could have its costs — that pro-AI forces would “try to destroy you,” according to one swing state Democratic political strategist granted anonymity to speak candidly."
• Vlasto and Moffat had colorful political histories before this role. Vlasto worked with Andrew Cuomo as "as an informal adviser to the governor as scandals over sexual misconduct engulfed him in 2020. During a 2021 deposition with the New York Attorney General’s Office, Vlasto was forced to recount an incident early on in Cuomo’s first term in which the governor called him an “incompetent asshole” in front of staffers."
• Vlasto rejects the idea that LTF's attacks on Bores helped him. Vlasto says of the idea that LTF's attacks actually helped Bores: "The idea that their money somehow hurt their reputation and elevated Bores? “If you ignore facts, numbers, polling and the ultimate outcome, if you set aside those things, then yes, I buy that theory,” said Vlasto." (Note: According to the NYT the ultimate victor in NY-12 Micah Lasher, who has perhaps even more stringent views on AI regulation than Alex Bores, told people during the campaign that he thought the LTF spending served to elevate Bores.)
• The article summarizes some of the most jaw dropping instances of astroturfing that LTF and its affiliate Build American AI engaged in - running a fake account pretending to be an AI safety proponent, and then having that account call for violence to stop AI development. "One such account, “@JonathanDoomer,” consistently posted memes with the kind of inflammatory anti-AI language that LTF frequently accuses its opponents of using. ... @JonathanDoomer posted pro-China content, anti-innovation content, and also trafficked in memes promoting violence in order to stop the spread of AI... Build American AI sought to downplay the effort, saying that it wasn’t a “core part of their strategy.”
• Zack Moffat expanded on the role that the astroturfing and sock puppet accounts in their approach: "“As we said, it wasn’t a core part of the strategy, it was tongue in cheek,” Moffatt added, referring to the meme sockpuppet accounts. “But we see ourselves as an omni-channel media operation that needs to get through to a variety of different audiences, both in Washington and Silicon Valley.”
• The article discusses the measures that OpenAI and Lehane took to distance themselves from LTF (Lehane was reportedly involved in helping to set up the organization). OpenAI and Lehane did not respond to a request for comment in the article about their connections to LTF.
I really couldn't get close to all the wild details in this post and its worth just reading the article in full (including a quote from yours truly). For those like myself who had been wishfully thinking that LTF might feel abashed after the Alex Bores campaign or revisit its decision to wage eternal war on "doomers" or regret/apologize for its absurd astroturfing campaigns, this article is about as strong a repudiation one can expect that the answer is no.\n\nQT @tcberenson: New POLITICO Mag cover story: no one is controlling more pro-AI money in the midterms than Josh Vlasto and Zac Moffatt at Leading the Future. @calder_mchugh sat down with them:
https://t.co/6XRpgpDFwB https://t.co/YP4BCSyMu9
See 1 related tweets
- @TaylorLorenz: RT @_NathanCalvin: This is a truly remarkable story that is worth reading in full for anyone who has...
18. daleverett (Group Score: 64.5 | Individual: 35.3)
Cluster: 2 tweets | Engagement: 10 (Avg: 16) | Type: Tech
If you’re building in this space, I’d like to help. Drop me a dm, I will fund your infra and make vc introductions. (I can refer you to finc as well)\n\nQT @daleverett: Building an agent startup? Especially company OS, brain OS, or AI teammates? These VCs have already backed the category 👇
• @USV — General Intelligence • @AcrewCapital — General Intelligence • @CompoundVC — General Intelligence • @LongJourneyVC — Eragon • @walden_catalyst — Nace • @generalcatalyst — Nace / Glean • @ICONIQCapital — TinyFish / Sierra • @USVP — TinyFish • @MongoDBVentures — TinyFish • @heavybit — Modiqo • @sequoia — Dust / Harvey • @kleinerperkins — Glean / Harvey • @a16z — Hebbia / Harvey • @IndexVentures — Hebbia
If you want to be best-in-class, use a backend that builds its own context layer for your ai agent/app
Sign up on https://t.co/Osu8LGArWL
(Dm me for free credits)
See 1 related tweets
- @daleverett: RT @daleverett: Building an agent startup? Especially company OS, brain OS, or AI teammates? These V...
19. ajambrosino (Group Score: 63.9 | Individual: 35.3)
Cluster: 2 tweets | Engagement: 330 (Avg: 364) | Type: Tech
👀\n\nQT @firesidealpha: Sam Altman reveals building Codex was a "crazy kamikaze mission" to beat Claude Code and that most of the best coders he knows now use Codex.
"We were way behind Claude Code, and it seemed like a kind of crazy kamikaze mission to try to beat them with a coding app."
"The consensus is that this kind of thing never works, and you just move on to the next one. It's a fool's errand to try to win when someone else already has momentum in a particular product category."
"But we decided we thought it was really important. We asked a team to do it, and they performed a legitimate, unbelievable, very rare in the history of business thing."
"Now it is the product that most of the best coders I know use."
"That felt like an impossible thing, but if we hadn't asked the team, hey, we have a really important but extremely hard mission for you, it just wouldn't have happened."
See 1 related tweets
- @madhavjha: RT @firesidealpha: Sam Altman reveals building Codex was a "crazy kamikaze mission" to beat Claude C...
20. BrianRoemmele (Group Score: 63.8 | Individual: 35.7)
Cluster: 2 tweets | Engagement: 44 (Avg: 244) | Type: Tech
By the way: This is the most Nvidia positive as one can get. This is not going to impact Nvidia sales, in fact if AI companies really understood what we are doing in the graves in the US and not what is taking place in China, they would hire us all, fast.
Nvidia GPU sales will soar and who adopts these garage method control the future of AI innovation.\n\nQT @BrianRoemmele: SUCCESS! AI ON NO GPU COMPUTERS!
I’m back the stress test is done.
AND I NOW KNOW I CAN SAVE LARGE AI COMPANIES BILLIONS OF DOLLARS!
I’ll show you how for free.
The 13B model ran fully offline on ordinary consumer hardware that has less RAM than the model wants.
Flash (the SSD) became the primary store.
Peak RAM stayed at 15.4 GB.
Average generation speed settled at 7.1 tokens per second once the working set was warm.
SSD bandwidth peaked at 412 MB/s during the heavier layers and dropped to a steady 180–220 MB/s while the narrative continued.
Coherence across multiple turns held—the same quiet storyteller voice that appeared on the $8 ESP32 simply scaled up. No cloud. No GPU. Just the design philosophy that constraints teach you more than excess ever will.
My goal is to have a full chat interface now working on it now!
I used the best-balanced 13B available for this exact job: Llama-2-13B-Chat Q4_K_M (7.48 GB on disk). It is the cleanest narrative model at this size, stays coherent when you feed it short personal notes, and refuses to invent worlds it cannot hold.
How you can test this on old hardware:
•Older Ryzen 5 desktop (any 6- or 8-core Zen 2/Zen 3 is fine) •24 GB DDR4 RAM (or less) •Single 1 TB NVMe SSD (PCIe 3.0 is what I used; PCIe 4.0 is faster but not required) •No discrete GPU!
Here is exactly how I set it up and how any amateur can repeat the experiment tonight on Linux, Mac, or Windows.
- Download the model (same on every OS)
one-time install of the downloader
pip install -U huggingface_hub
pull the exact file I used
huggingface-cli download TheBloke/Llama-2-13B-chat-GGUF
Llama-2-13B-chat.Q4_K_M.gguf
--local-dir ./models
- Build / install llama.cpp Linux git clone https://t.co/DQz9eCwnz0 cd llama.cpp mkdir build && cd build cmake .. -DLLAMA_NATIVE=ON cmake --build . --config Release -j$(nproc)
macOS (Apple Silicon or Intel) git clone https://t.co/DQz9eCwnz0 cd llama.cpp mkdir build && cd build cmake .. -DLLAMA_METAL=ON -DLLAMA_NATIVE=ON cmake --build . --config Release -j$(sysctl -n hw.ncpu)
Windows •Download the latest release zip from https://t.co/J6YTyUU81F •Or build with Visual Studio: git clone https://t.co/DQz9eCwnz0 cd llama.cpp mkdir build cd build cmake .. -G "Visual Studio 17 2022" -A x64 cmake --build . --config Release The binary ends up in build\bin\Release\llama-cli.exe.
- Run the exact command that produced the numbers above Linux / macOS ./build/bin/llama-cli
-m ./models/Llama-2-13B-chat.Q4_K_M.gguf
--mmap
-c 4096
-n 512
-t 6
--temp 0.65
-p "You are a personal knowledge node. Here are today’s notes and schedule. Weave them into a quiet narrative with a beginning, middle, and gentle intention."
Windows .\build\bin\Release\llama-cli.exe -m .\models\Llama-2-13B-chat.Q4_K_M.gguf --mmap -c 4096 -n 512 -t 6 --temp 0.65 ` -p "You are a personal knowledge node. Here are today’s notes and schedule. Weave them into a quiet narrative with a beginning, middle, and gentle intention."
That is the full setup. The --mmap flag is the entire secret: the weights stay on the NVMe, the OS pages in only the active tensors, and the sparsity windowing I added on top simply keeps the most recent neuron blocks hot.
7.1 tokens per second on a machine that most people would call under-powered.
Peak RAM never touched the 24 GB ceiling.
The narrative never broke.
This is no longer a toy. It is a personal knowledge node that lives entirely offline, owns its own data, and costs almost nothing to run.
The same design that made the $8 ESP32 speak now speaks on every desk that already has an SSD.
Soon we will explore this Flash setup on mor elaborate chat type interfaces.
But the big story is how this will save RAM, power and heat at very large data centers. My goal is to bypass CUDA on some tasks and remove the need for RAM.
1 of 2
See 1 related tweets
- @BrianRoemmele: RT @BrianRoemmele: SUCCESS! AI ON NO GPU COMPUTERS!
I’m back the stress test is done.
AND I NOW KN...