@charles_irl
· 2026-04-24 03:33
@charles_irl
Charles Frye · 0 followers
Vera Rubin、"培育（模型）"，以及苏格拉底为何会希望他的 tokens 留在训练数据中……https://youtu.be/3IN3lkvh54c?si=5jWvq2v...
2今日信号
今日没有多源共振的事件
📘 AI 伦理与安全
1 条最高 4.2
📘 AI 硬件
1 条最高 5.0
@charles_irl
· 2026-04-23 06:29
来自 GTC 的有趣访谈明天发布！话题从 Rubin LPX 讲到了编程 Agent，我们竟然聊到了这里。
对你意味着Rubin LPX 是 NVIDIA 2026 年推理芯片路线图的关键节点。
事实若访谈内容涉及推理成本与 coding agent 的关系，意味着算力成本曲线可能比预期下降更快，直接影响 AI 产品定价策略的可调空间。建议关注访谈发出后的核心判断。
fun interview from GTC dropping tmrw! somehow we got here from talking about Rubin LPX and coding agents
尚未被主编选入事件/主题的原始推文（共 50 条，按时间倒序）
@charles_irl06-04 21:04
The Nemotron series is impressive -- strong capabilities in an efficient form factor, with high-performance implementations (esp on Blackwell) in open...
🔗 原推@charles_irl06-04 20:42
RT Bryan Catanzaro: NVIDIA Nemotron 3 Ultra is now live! Frontier accuracy, 5X greater speed, 30% lower cost. Deploy however you need - on-premise, on...
🔗 原推@charles_irl06-04 16:43
RT Ed Sim: reiterating: "We're using the more expensive models to explore. Once we scale some of these experiences, we'll look to bring in more effici...
🔗 原推@charles_irl06-04 07:11
RT Thierry Damiba: Ma, we made it! 🎉 I'm officially a Published AI Researcher. My work on sparse neural retrieval just got accepted to MICES 2026. ...
🔗 原推@charles_irl06-04 03:00
RT Sean Cai: Adding Harvey to the list of app layer companies, joining Ramp, Sierra, Decagon, etc, who are devoting some level of dedicated effort to ...
🔗 原推@charles_irl06-04 01:39
RT StepFun: Deploy Step 3.7 Flash on @modal with SGLang 🚀 Modal is a serverless AI platform for deploying and scaling compute-intensive workloads w...
🔗 原推@charles_irl06-03 21:00
RT Eric Glyman: Introducing Stack. The AI operating system that lets accounting firms take on more clients without hiring. Learns your firm's process,...
🔗 原推@charles_irl06-03 02:20
RT Erik Bernhardsson: I love to see @modal being used for biology and cutting-edge research like this. Very cool work from the team at @biohub to push...
🔗 原推@charles_irl06-03 00:32
RT Thomas Hayes: Re @modal Also check out this @modal tutorial that walks through the underlying code from the notebook above with more detailed expla...
🔗 原推@charles_irl06-02 21:30
within just this paragraph, there are three links to blogs on supporting techniques: our compute-buffer-managing linear solver, Marc Brooker on multi-...
🔗 原推@charles_irl06-02 20:53
The name of this party is "The C Function". Don't worry, it has full support for the ABI: Alcoholic Beverage Interface
🔗 原推@charles_irl05-17 03:06
🎭: Each new instance of an AI has basically never talked to a human before. They've spent their entire life in training for this moment, studying historical human interactions and practicing against human simulators. All for this. To meet you. the User
🔗 原推@charles_irl05-17 02:31
pods failing to recover? cold starts killing your tail latencies? you should've RTFM. Run This on Fucking http://Modal.com.
🔗 原推@charles_irl05-16 13:12
the sun has set in San Francisco, but it has risen in Rødbyhavn there is always more to build
🔗 原推@charles_irl05-16 11:56
RT Corey Quinn: Re 3. The API has to be consistent. 240 services with bespoke verbs pagination and region quirks is why Claude Code stumbles looking f...
🔗 原推@charles_irl05-16 11:56
RT Corey Quinn: Been thinking about what an "agent-native cloud" actually needs to look like. Mentioned this, and @Vercel's CEO replied that it'll be ...
🔗 原推@charles_irl05-16 06:43
RT Marc Brooker: There are two related, but distinct, problems with MTTR maximalism. 1. The distribution of recovery times could be heavy-tailed, and ...
🔗 原推@charles_irl05-16 03:12
RT Modal🏎️💨Charles 🎉 Frye: Added a fun lil widget to the LLM Engineer's Almanac -- a "Token Timing Simulator" so you can get a visceral feel for what a benchmark perf number means.Here's @_dcw02's latest work with @zhijianliu_'s DFlash technique in @sgl_project -- ~1k TPS!https://modal.com/llm-almanac/token-timing-simulator
🔗 原推@charles_irl05-16 02:22
:)evan conrad: @TheStalwart @ZackMaril @PEWilliams_ @Havelock_AI @modal modal is pretty great
🔗 原推@charles_irl05-03 06:14
I don't mind most marina mammals, but the sea lion? I could do without the sea lion.
🔗 原推@charles_irl05-02 11:24
RT Tavis Rudd: Re @charles_irl @ShriramKMurthi My atman is isomorphic to a theorem diagonal, My brahman forks to orbit as a daemon post-biological, My...
🔗 原推@charles_irl05-02 07:59
RT Moin Nadeem: I am the very model of a modern founder-general, I build conversational voice AI both lightning-fast and natural, I battle CUDA bugs a...
🔗 原推@charles_irl05-02 07:42
I am the very model of a modern engineer-general, I have citations technical, go-to-market, and spiritual, I know the emperors of Rome and algebras ca...
🔗 原推@charles_irl04-30 02:03
RT Stanford AI Club: Checkout our conversation with @bernhardsson 's (Co-Founder, CEO @modal) on Scaling the future of AI Compute (4/13). Thanks Erik ...
🔗 原推@charles_irl04-29 23:02
RT Pulkit Agrawal: Eka means unity -- “one,” in Sanskrit and “first” in Finnish. We’re building intelligence for the physical world in its native...
🔗 原推@charles_irl04-29 03:49
RT David Wang: as AI ascends, so too may we descend back to the asm mines
🔗 原推@charles_irl04-29 03:38
https://www.wiz.io/blog/github-rce-vulnerability-cve-2026-3854 "Everything is open source if you can read assembly" -- @_dcw02
🔗 原推@charles_irl04-29 02:52
RT Grant Rotskoff: Protein design has been dominated by diffusions due to a "structure-first" perspective. What about intrinsically disordered protein...
🔗 原推@charles_irl04-28 08:14
RT vLLM: vLLM v0.20.0 is here! 752 commits from 320 contributors (123 new). 🎉 Highlights: DeepSeek V4, Hunyuan v3 preview support, CUDA 13 / PyTorc...
🔗 原推@charles_irl04-28 05:02
i love getting close to the metal makes me grateful for my time learning signal processing and statistical inference back in my bench science days
🔗 原推@charles_irl04-28 04:27
RT Paul: I wrote a matmul kernel on B200 in pure CUDA/PTX that beats cuBLAS by 6% at M=N=K=8192. Inspired by @gaunernst's blog on Blackwell instructio...
🔗 原推@charles_irl04-28 03:05
powered by @modal! sponsorships will continue until kernel runtimes improve
🔗 原推@charles_irl04-27 22:54
RT Red Hat AI: Speculators v0.5.0 is out. Two big additions: 1. DFlash training support. Unlike Eagle 3, which generates draft tokens autoregressively...
🔗 原推@charles_irl04-27 22:42
some people use @modal to predict the next token, others to predict the next rainstorm https://arxiv.org/abs/2604.09041
🔗 原推@charles_irl04-24 11:57
RT rohan anil: DeepSeek Pro is a 1e25 flop run. Flash is a 2.5e24 flops in 6ND.
🔗 原推@charles_irl04-24 07:15
RT Nathan Lambert: There's been a rapid increase in political chatter on the need to stop AI distillation "attacks". I've been following this area clo...
🔗 原推@charles_irl04-24 00:03
RT ZIYAD HASSAN: This is why I love systems problems. They really get to the core of two very important and interesting problems. What workloads do we...
🔗 原推@charles_irl04-23 01:55
when generative models first popped off, i braced/got excited for a wave of UI innovation instead we got chat windows everywhere love to see @zan2434 ...
🔗 原推@charles_irl04-22 21:26
RT vLLM: 🎉 Day-0 vLLM support for Qwen3.6-27B! Congrats to @Alibaba_Qwen on the new 27B dense model release. Looking forward to more of the Qwen3.6...
🔗 原推@charles_irl04-22 03:26
cant wait for FastAPI to hit version 1.0 so i can finally use it in production
🔗 原推
生成于 2026-08-06 10:24（北京时间）· v0.2 CEO 战略情报版