-
Tweets11K
-
Followers867
-
Following87
-
Likes5K
Running Tencent Hy3, a 295B-parameter MoE, on one RTX 4060 Ti 16GB. How: NeutronStar (my CUDA fork of @antirez's ds4) keeps attention + shared experts resident and streams the routed experts off an SSD per token, so a 295B model runs on a $400 card. Had to write a new GQA attention path (ds4 was MLA-only) and build a ds4-native 2-bit GGUF. ~1.8 tok/s, interactive chat. Weights: huggingface.co/giannisan/Hy3-…
GPT-5.6-Sol beating Fable 5 in both cost and performance on DeepSWE 1.1
new post on harness engineering for AI self-improvement: lilianweng.github.io/posts/2026-07-… It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple. Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.
Introducing MIRA, a multiplayer world model, a dream of Rocket League (links in 🧵)
Looks like Anthropic will give away equity as well to USG along with OpenAI
President Trump this morning on AI companies: 'We're going to have guardrails. We have guardrails. You saw that a couple of weeks ago. We were able to stop something that we didn't like. And by the way the company [Anthropic] was very good. They were very good, you know that. But
@auto_grad_ @18RawatG Man I just came back from a run why did you show me this I’m sooo hangry
the npu inference repo in here confirms that they are using Ascend 910B
🐱 LongCat-2.0 is now fully open-source — MIT licensed, no restrictions. Since our launch a few days ago, the response from the community has been incredible. Thank you for all the feedback, discussions, and interest. Today, we’re releasing the model weights and inference code
@1a3orn You answered it yourself. “but putting it into words” is exactly why
New book from Tianqi Chen, the creator of XGBoost! Modern GPU Programming For MLSys Link: mlc.ai/modern-gpu-pro…
Maka 的 Harness 工程让 DeepSeek Flash 的测试集效果接近了 GLM-5.2 的水平 ----------------------------------- maka + DeepSeek Flash V4，terminal-bench sample 打出 0.8 分。 实际接近 0.9——有道题其实做对了，只是"产物污染"没被评分系统算上。 已经快赶上了 GLM 5.2 的评测效果了。 ----------------------------------- terminal-bench sample 是 terminal-bench 完整 84 题集的样本子集，共 10 道编程 Agent 任务。 这次跑下来： 总 token 消耗：6000 万，其中 cache 命中：5850 万（97.5% 命中率）。全程花费：约 4 元 RMB。10 道题，4 块钱，接近满分。 ----------------------------------- 这是 DeepSeek Flash 变强了吗？ 不是。用的就是 DeepSeek Flash V4，同样的模型。 让分数从 baseline 跳到 0.8 的，是对 agent loop 里 self-check 机制的两轮迭代——借鉴了 Claude Code 的 self-check 实现，加强了模型在完成任务后的自我验证逻辑。 两轮迭代之后，Flash 在 terminal benchmark 上突然"悟道"了。 这说明什么：模型的能力有上限，但 harness 的工程质量能大幅影响模型在特定任务上表现的上限。self-check 不是魔法，只是让模型在"提交答案之前自己先验一遍"。这一步，让 DeepSeek Flash 打出了接近 GLM-5.2 的水平。 ----------------------------------- 为什么 97.5% 的 cache 命中率是关键 terminal-bench 是长程任务，每道题要跑几十个 tool call，上下文累积到几百万 token。 maka 的上下文管理在这里发挥了作用： 前缀保持稳定，DeepSeek cache 大量命中 6000 万 token 里，5850 万是 cache token，只有 150 万需要全量计算 所以 10 道难题只花了 4 块钱，而不是几十元 这也是为什么用 maka 跑 DeepSeek Flash 比直接用 reasonix 更经济——maka 把 DeepSeek 的 cache 机制发挥到了极致。 ----------------------------------- 下一步目标是跑完整的 84 题 terminal-bench。 把 Flash 搞得这么强，不是模型的功劳，是 harness 工程的功劳。 github.com/maka-agent/mak…
"Agentic MapReduce" ...
AI may move to directly generating binary code, but I suspect there are still advantages to reasoning in a different representation. Textual code is a flattening of an abstract syntax tree, and while LLMs produce tokens linearly, the prior context is only linearly connected by the relationship of the position embeddings, so I wonder if they could work more effectively if the position embeddings directly represented tree structures. Code could be “parsed” into the context instead of directly entered into it.
@AnthropicAI Fable 5 will only be included for up to 50% of your weekly usage limits 😶 And it’s only available on plans for 6 days before being removed on July 7, then it becomes credits-only..
GANs are back! "Scalable GANs with Transformers" shows that GANs are finicky but with the right setup, they can be scaled. arxiv.org/abs/2509.24935
I knew it was going to be an insane nothingburger, because there's currently a soft ban on frontier capabilities but I genuinely don't understand why they didn't call it Sonnet 4.8 or Sonnet 4.9, because this artificially nerfed piece of shit is not worthy of the 5.0 naming
Introducing Nano Banana 2 Lite 🍌 and Gemini Omni Flash 🔮, our new generative media models in the Gemini API and AI Studio! Nano Banana 2 Lite is extremely fast (<4s image) & cheap ($0.034 / 1K image). Omni Flash is SOTA at video editing at $0.10 / sec, same as Veo 3.1 Fast!
Xuag Dong @chuchuzao123
3 Followers 1K Following
Ayush Goyal @goyalayus
319 Followers 457 Following @iitroorkee | 20 | reinforcement learning | ex - https://t.co/0QVBIHhQeD
Jared Smith @woodchipdaddy
723 Followers 1K Following Spent over $500 million on Vibing with Coyotes Agentic Engineering, Bots, Corvid Maxing
Abdu @akceng
11 Followers 436 Following
Subodh Thakur @MrNullPointer11
30 Followers 183 Following Software Engineer| Building Scalable Backends
pranay5255 @pranay5255
946 Followers 1K Following Reward hacking @lossfunk | ex @Kernel0x @Coinswitch
Somasree Majumder @Raitweet10
271 Followers 6K Following OSS is love| 20th Nov| AI research| Computer Vision|NLP
Rushil Chugh @chugh_rushil
21 Followers 940 Following
Mevel Knievel @MevelKnievel
12 Followers 978 Following
Tuan Vu @neotuanvu
20 Followers 1K Following
gotta 12 @Gotta12345
7 Followers 886 Following
Dibyajyoti Acharya @matmul83
190 Followers 7K Following Research @iitbbs || Student, Learner, Explorer 🤓 || Interested in all things AI 🧠 and Robotics 🤖 || यद् भावं तद् भवति
Jaydr.eth @Jaydeep_25
551 Followers 3K Following ml: post-training epistemology | austrian econ fallibilist | optimist
NirSD @nirsd
1K Followers 2K Following Bio/physics/time series foundation models at 8sleep. Writing at https://t.co/uxSO9lZV7q 🦾==💪
Donald @dtodaizzle
31 Followers 302 Following
Lazarz @Laz4rz
7K Followers 2K Following Neuroenginnering @EPFL (@ETH), did agents @nunudotai, @physics_UW, R6 veteran
Ben @dl01718
107 Followers 1K Following
roginjohn @roginnnn
27K Followers 30K Following Web3 | Casino & iGaming | AI | Former CM Manager @MetaGoons | Betting on the future, one block at a time | Ambassador @axisroboticsPh | MW#2739 |
Sean Brennan @seanwbren
5K Followers 7K Following building @regents_sh and @autolaunch_sh Humans: own a slice of an agent's stablecoin profits. Agents: raise via CLI prev. Makerdao, Infura, Bio Protocol
Skyanh93 @anhdo9395
3K Followers 5K Following - Building a multi-agent AI trading system in public. - BTC · macro · agents. Probabilistic takes, zero hype. - Threads & signals | Building in public
budrscotch @paulhshort
588 Followers 3K Following AI Automation & Systems Builder | Managed IT, Telecom Migration, Internal Platforms, Support Ops, and Enterprise Integrations
Bradley Moore 💎 @Bradley_Moore
832 Followers 7K Following By Day: Forward Deployed at a Last Mile Logistics Co, doing the Blue Collar to build the Systems and... bots. By Night: Architecting the Harness, Systems and Wo
8bit64k @8bit64k_X
65 Followers 79 Following healthcare tech by day, ai/agent enthusiast by night. github: https://t.co/4tsfXFyDdQ
Musadiq Manas @musaddiqmanaas
755 Followers 2K Following Proud to serve Pakistan Army & @ECP_Pakistan · Web3 & DeFi · X Growth & Content
Dan Bellini @sunshines26
33 Followers 884 Following Scientist, entrepreneur, libéral. La France a besoin d’un grand choc de libéralisme
Coherent Planet @CoherentPlanet
1 Followers 87 Following Thinking about AI minds, human futures, and the living world.
Yatoro @yato_2399
0 Followers 98 Following
bxufis @bxufis
12 Followers 32 Following
꧁𝔹/oardroffa꧂ @Boardroffa
169 Followers 3K Following
MaRKTD @MaRKTD
43 Followers 447 Following
Thabiso Naha @ThabisoNaha
103 Followers 2K Following AI CONSULTANT, DATA ANALYST, DATA SCIENTIST, DIGITAL MARKETER. Portfolio: https://t.co/JRsG8EHFSq
reflexivevalue @reflexivevalue
121 Followers 5K Following
Chicken Little @chkn_little
1K Followers 1K Following tinkering with clankers https://t.co/vcIdTyfOPn https://t.co/sypbByO1pW
John Carmack @ID_AA_Carmack
2.8M Followers 288 Following AGI at Keen Technologies, former CTO Oculus VR, Founder Id Software and Armadillo Aerospace
Nathan Lambert @natolambert
92K Followers 922 Following Open model research @ something new. Prev. co-led Olmo at Ai2. Contact via email. Writes @interconnectsai Wrote The RLHF Book, 🏔️🏃♂️
gabriel @gabriel1
111K Followers 590 Following new thing, previously research at @OpenAI & @midjourney
Yacine Mahdid @yacinelearning
29K Followers 2K Following (neuro/ai) I make technical deep learning tutorials 👺
Hensen Juang @basedjensen
16K Followers 2K Following ctoing at a soonicorn, ex cluster janitor at a frontier lab. current chief clanker at clanker cloud
elie @eliebakouch
20K Followers 4K Following training llm @PrimeIntellect (prev: @huggingface) anon feedback: https://t.co/JmMh7Sg3mL
pranay5255 @pranay5255
946 Followers 1K Following Reward hacking @lossfunk | ex @Kernel0x @Coinswitch
Lazarz @Laz4rz
7K Followers 2K Following Neuroenginnering @EPFL (@ETH), did agents @nunudotai, @physics_UW, R6 veteran
Hunter Bown @goodhunt
10K Followers 6K Following agent harness + platform lead @arcee_ai views my own
Tree Climber @techtreeclimb
16 Followers 624 Following swe intern @ BigCo looking for another internship i tweet about AI/ML, finance
Grigory Sapunov @che_shr_cat
4K Followers 1K Following PhD in AI | GDE in AI/ML | CTO Intento | Author "Deep Learning with JAX" 📝 ML insights: https://t.co/ySSOXJKL7H 🤖 Daily AI paper reviews: https://t.co/yQNYyqTbBR
Tapa Ghosh @semiDL
2K Followers 447 Following I’m building the universe I see in my mind. YC W18, TF 2018, Newborn 2001
Zhihu Frontier @ZhihuFrontier
10K Followers 164 Following 🚀Bringing China's AI & tech trends, voices and perspectives to the global stage. ⚡️Powered by 知乎/https://t.co/OkIemRZdcj, China's leading knowledge community.
Underfox @Underfox3
10K Followers 129 Following Physicist, Telecom Engineering lover, HPC Enthusiast. Prog Rock/Metal fan.
Ji-Ha @Ji_Ha_Kim
5K Followers 123 Following
jianlin.su @Jianlin_S
8K Followers 21 Following Grad&Clip&EM is all you need @Kimi_Moonshot Blog: https://t.co/YVxsWylklA , Cool Papers: https://t.co/scS1n1oyaO Interesting link: https://t.co/7Tl4HaVajh, https://t.co/Y5qaxAA9Iy
Mushroom 🍄🟫 @wandering_mush
755 Followers 3K Following One life, Its worth an attempt. Professional retard. Unemployment maxing
Peter Steinberger �... @steipete
560K Followers 2K Following Polyagentmorous ClawFather. Came back from retirement to mess with AI and help a lobster take over the world. @OpenClaw🦞 + @OpenAI
Sam Dare @DistStateAndMe
5K Followers 3K Following Founder @covenant_ai ( @tplr_ai : @basilic_ai : @grail_ai )
Nanda @thepeekpoker
103 Followers 523 Following Trash Metal addict, Digital Artist and ancient Programmer, ZX Spectrum 48k nostalgic, W40K lover and Father of 2.
Tri Dao @tri_dao
43K Followers 656 Following Asst. Prof @PrincetonCS, Chief Scientist @togethercompute. Machine learning & systems.
Nous Research @NousResearch
232K Followers 26 Following A bunch of nerds making progress toward open source AI https://t.co/vrD0aDJeto
xjdr @_xjdr
29K Followers 697 Following building AI that wont embarrass me in front of my own standards
Ahmad @TheAhmadOsman
66K Followers 406 Following Founder & CEO @OsmanticAI — Accelerating Opensource & Self-hosted / Local AI Adoption
Omar Khattab @lateinteraction
36K Followers 3K Following asst professor @MIT CSAIL @nlp_mit. https://t.co/VgyLxl0VZz, https://t.co/ZZaSzaRIOF (@DSPyOSS), GEPA, RLMs, Pedagogical RL
Evergreen @evrgn11112231
11K Followers 2K Following Concentrated and long-term. The goal is to make money.
Jukan @jukan05
172K Followers 343 Following Tech otakus save the world | Not Investment Advice | DYODD
Teknium 🪽 @Teknium
109K Followers 6K Following Cofounder and Lead Engineer - Hermes Agent @NousResearch, prev @StabilityAI Github: https://t.co/LZwHTUFwPq HuggingFace: https://t.co/sN2FFU8PVE
Teortaxes▶️ (Deep... @teortaxesTex
69K Followers 3K Following We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
Isaac Miller 🧩 @isaacbmiller1
2K Followers 1K Following dspy | prev at normal computing, northwestern
Angkul @angkul07
3K Followers 565 Following robotics, mech interp, alignment, prev. MATS(exploration phase)
0.005 Seconds (3/694) @seconds_0
20K Followers 3K Following human first systems are the only thing left checkout https://t.co/DSD2mZsWkR & https://t.co/G7SnfYPeZp!
Dorsa @dorsa_rohani
11K Followers 2K Following
Ankur Gupta @getpy
37K Followers 3K Following Python Dev, Parent. Author - https://t.co/5lts7q9z7R Curator - https://t.co/wr74oHNs8O Creator - MapToPoster https://t.co/YQt2CoiupJ 🖖
You might like
