ML @ FAIR, ex-Uber AI Labs, ex CloudFlare SAM, Perception Encoder, PerceptionLM, POET... Author of https://xitter.cf/t.co/rwle5Fc66f Blog: https://xitter.cf/t.co/25zHKG86y4
San Francisco, CA
Joined April 2009
- Tweets234
- Following632
- Followers261
- Likes466
Pinned Tweet
Proud to have contributed to SAM 3! Good work team 👏 Super excited to see people using it and what you'll build with the new capabilities.
Check out the demo: segment-anything.com
Today we’re excited to unveil a new generation of Segment Anything Models:
1️⃣ SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts.
🔗 Learn more about SAM 3: go.meta.me/591040
2️⃣ SAM 3D brings the model collection into the 3rd dimension to enable precise reconstruction of 3D objects and people from a single 2D image.
🔗 Learn more about SAM 3D: go.meta.me/305985
These models offer innovative capabilities and unique tools for developers and researchers to create, experiment and uplevel media workflows.
Jiale Zhi retweeted
We’re releasing SAM 3.1: a drop-in update to SAM 3 that introduces object multiplexing to significantly improve video processing efficiency without sacrificing accuracy.
We’re sharing this update with the community to help make high-performance applications feasible on smaller, more accessible hardware.
🔗 Model Checkpoint: go.meta.me/8dd321
🔗 Codebase: go.meta.me/b0a9fb
Jiale Zhi retweeted
@bcaine @officialLoganK This chip board test is fun! SAM 3 definitely cannot miss it. I used "port", "capacitor", "CPU socket" to call SAM3, and got the results! Better than both GPT 5.2 and Gemini 3. Maybe it's time to distill SAM 3
#SAM3 #GPT52 #Gemini3
It is a bit concerning that GPT-5.2 has the widest gap between reported SWE-bench verified number and mini-swe-agent reported number.
Updates to official SWE-bench leaderboard: GPT 5.2 high is #3 after Gemini for similar price point. GPT 5.2 medium reduces gap to Sonnet 4.5 and is much cheaper. All independently evaluated using identical mini-swe-agent setup. More info & full trajectories in 🧵
Bump into @hardmaru randomly at NeurIPS, nice to catch up and congrats on @SakanaAILabs series B!
True, but I would argue that the capability required to understand some of the django tasks are transferable. Navigate through ambiguous user requirements, dig deep into the codebase to locate the issue, fix it, test the fix are all required to get good performance. For anyone who has a doubt on this, try to solve this yourself tbench.ai/registry/swebench-…
Claude Opus 4.5 claims state-of-the-art performance on SWE-bench verified. Interestingly, they no longer report results with parallel test-time compute, even though they used it when comparing against human candidates. Is additional compute just not needed anymore?
This reminds me of Steve Jobs' comment on "taste" youtube.com/watch?v=5y03eFMm…
A few interesting observations:
1. Tried google's Antigravity for Gemini 3.0 Pro initially, but the rendered the image is pure black, switched to cursor then it becomes normal
2. Despite the image issue, Gemini 3.0 in Antigravity is the only one that read the final image for validation.
3. Even thought I mentioned top lighting in the prompt, composer-1 gets the scene upside done. This is something I haven't noticed for other models
Jiale Zhi retweeted
Note: I'm actively looking for a position (at any level) to pursue work in this domain (meta-learning, plastic networks, and other neuro+AI topics).
If you're aware of any openings, let me know!
New paper!
We meta-train plastic networks to do transitive inference (learn A>B & B>C, infer A>C)
Training discovers a new neural learning algorithm, that reinstates past items in memory to quickly reconfigure existing knowledge upon learning a new item
biorxiv.org/content/10.1101/…
Jiale Zhi retweeted
✨NEW LAUNCH! LLaMA2 chat API & open-source playground💫:
We're releasing tools that make it easy to test @meta's latest LLM & add it to your own app with @replicatehq.
Playground: llama2.ai
Live chat API here: replicate.com/a16z-infra/lla…
Repos & instructions below:
Last weekend, I went to Berkshire Hathaway's annual meeting and was very lucky to a chance to ask Warren Buffett and Charlie Munger a Question, here is my experience:
linkedin.com/pulse/i-asked-w…