CHAI / William Beauchamp. (@chai_research) header
CHAI / William Beauchamp. (@chai_research) profile picture

Founder of CHAI - Building social AI platform in Palo Alto. ✝️

Palo Alto, CA
Joined May 2023
CHAI / William Beauchamp. retweeted
App Review: 375k apps a week on a team built for 125k. Historically, half of all submissions are resubmissions. Apple blames AI. I broke the two habits behind this in 2009. They’re back. phillipshoemaker.substack.co…
4
2
1
38
6,234
Anyone seeking to understand Apple App Review should start with @pbsIdentity Great article this week on account termination - phillipshoemaker.substack.co…
327
2
2
37
70,758
Great article from @pbsIdentity - Changing Leadership at Apple and App Store substack.com/@phillipshoemak…
513
1
6
31
118,726
If anyone needs advice on App Store rules and policy - this Substack is treasure trove of gems and wisdom. phillipshoemaker.substack.co…
170
2
1
38
92,062
This technique is how startups without billions in funding are able compete with frontier labs. Black-Box On-Policy Distillation of Large Language Models Qwen 14B (Student) out performing GPT-5-Chat (Teacher) arxiv.org/pdf/2511.10643 AMD - MI325X is still a beast, with 256GB of VRAM - can train huge sota models fast 🥇
670
4
13
105
290,084
CHAI just hit $70M in ARR with only 11 engineers. Small team. Iterating fast. Shipping multiples time a day. Taking big risks. Hiring in Palo Alto.
1,064
19
13
425
450,978
CHAI / William Beauchamp. retweeted
Speed is the moat. MI455X is both fast, and is the fastest I have seen in execution for bringup of a complex GPU platform. MI455X is right on target for shipments in 2H2026 - Irrespective of what @SemiAnalysis_ says - rev your engines because speed is coming.
38
58
25
572
242,343
Amazing work from the team in 2025. $50M in revenue with only 10 engineers. Incredibly proud of you guys! ❤️
291
6
8
135
206,463
Which do you prefer
59%Test Driven Development
41%Vibe Coding
99 votes • Final results
66
1
56
146,192
Well done! 👍
CoreWeave benchmarks the NVIDIA GB300 NVL72 📈 >6x higher throughput per GPU vs. H100 Read more about this breakthrough in our new blog → hubs.la/Q03FDFZF0 #GB300 #Deepseek #CoreWeave #AI
6
37
148,396
A noval neural architecture that aims for efficiency. 👏 It auto-replaces less-useful transformer layers into linear attention layers which reduces KV-cache and speeds up compute. 53x faster generation and 6x prefilling.🚀🚀 🚀
9
2
33
90,916
Jet-Nemotron: Efficient Language Model with Post Neural Architecture Search arxiv.org/pdf/2508.15884v1
4
9
63,784
AMD x IBM Quantum x AI 🔥🔥🔥
AMD and @IBM are partnering to build hybrid quantum-centric supercomputing, fusing quantum and AI with the best of high-performance hardware and software to solve the world’s toughest real-world challenges at unprecedented speed and scale. amd.com/en/newsroom/press-re…
2
1
14
65,646