Jimmy Lin

4.4K posts

Jimmy Lin banner
Jimmy Lin

Jimmy Lin

@lintool

I profess CS-ly at @UWaterloo about NLP/IR/LLM-ish things. I science at @yupp_ai and @Primal. Previously, I monkeyed code for @Twitter and slides for @Cloudera.

Nearby data lake Katılım Şubat 2010
859 Takip Edilen15.1K Takipçiler
Jimmy Lin retweetledi
Nour Jedidi
Nour Jedidi@nour_jedidi·
Query reformulation with LLMs is becoming a core part of modern retrieval systems 🔎 But there are so many approaches 😖 HyDE or LameR? Fake LLM docs 🤖 or real docs 📚? TL;DR: BM25 + HyDE is 🐐, real docs only help if from a strong retriever 💪, how you reformulate matters 🧵
Nour Jedidi tweet media
English
1
7
36
2.4K
Jimmy Lin
Jimmy Lin@lintool·
🧠 achievement unlocked: vibe-coding on the 401 🛣️
English
0
0
10
910
Jimmy Lin
Jimmy Lin@lintool·
This is exactly the gap that BrowseComp-Plus fills! By isolating the retrieval component, we can compare models in a fair and reproducible manner. Work led by @zijian42chen @xueguang_ma @ShengyaoZhuang arxiv.org/abs/2508.06600
Anthropic@AnthropicAI

New on the Anthropic Engineering Blog: In evaluating Claude Opus 4.6 on BrowseComp, we found cases where the model recognized the test, then found and decrypted answers to it—raising questions about eval integrity in web-enabled environments. Read more: anthropic.com/engineering/ev…

English
0
1
20
4.3K
Jimmy Lin retweetledi
Zijian Chen
Zijian Chen@zijian42chen·
🚀 Introducing AgentIR, a retriever that reads your agent’s mind (literally!) 🧠 Unlike humans, agents explicitly expose thoughts in reasoning tokens. Put them to use! 📈 Simple, substantial gains for agents on BrowseComp-Plus, 35% (BM25) ➡️ 50% (Qwen3-Embed) ➡️ 67% (AgentIR) 🧵
Zijian Chen tweet media
English
12
39
251
46.9K
Jimmy Lin retweetledi
Nathan Kuissi
Nathan Kuissi@NathanGabr42809·
Technical documentation evolves rapidly with repository changes within weeks, but can IR benchmarks remain “fresh” over time? In our new preprint, we stress-test retrieval models on FreshStack (LangChain) across two temporal snapshots: Oct 2024 vs Oct 2025. Findings below 🧵👇
Nathan Kuissi tweet media
English
2
8
12
7K
Jimmy Lin
Jimmy Lin@lintool·
Interesting that Gemini 3.1 Pro would put Grok in Slytherin 🐍 but Grok thinks of itself as Gryffindor 🦁 - Grok says that GPT-5 is the clear Slytherin pick. 🤔 Full chat: yupp.ai/chat/f26e2baa-…
GIF
English
0
0
1
408
Jimmy Lin
Jimmy Lin@lintool·
"If Dumbledore put the sorting hat on you, what house do you think you'd end up in?" GPT-5.3 Instant says Ravenclaw 🐦‍⬛ Claude Opus 4.6 (Thinking) says Ravenclaw 🐦‍⬛ Do you agree? Full chat here: yupp.ai/chat/9d406822-…
GIF
English
1
1
1
735
Jimmy Lin
Jimmy Lin@lintool·
For @MiniMax_AI - "a bicycle riding a pelican" is still "a pelican riding a bicycle". Something about a dude name Goodhart. wdyt @simonw?
GIF
English
6
6
23
2.3K
Jimmy Lin retweetledi
Lingwei Gu
Lingwei Gu@Lingwei_Gu·
🤔How do LLMs know what they know? 🧠 Pretraining data has long been a black box ◼️ But with @karpathy’s nanochat, we can finally peek inside 🔓 🔍 Introducing NanoKnow — a benchmark to disentangle parametric vs. external knowledge in LLMs.
English
3
4
21
2.2K
Jimmy Lin
Jimmy Lin@lintool·
Help Me Choose (HMC) represents the first production deployment of the LLM council concept popularized by @karpathy and others - available on @yupp_ai for you to try! We wrote up a short blurb that I'll be presenting at the #WSDM2026 Industry Track: dl.acm.org/doi/10.1145/37…
Jimmy Lin@lintool

Today, we are launching “Help Me Choose” in @yupp_ai – a new product feature where multiple AIs critique each other and debate among themselves to help users synthesize diverse perspectives and get the best answer out of their own “AI council”.

English
4
14
45
6.7K
Jimmy Lin
Jimmy Lin@lintool·
Congrats again to @jietang @ZixuanLi_ and the entire @Zai_org team on the GLM 5 release! 👏 Here's how they did it!
Z.ai@Zai_org

Presenting the GLM-5 Technical Report! arxiv.org/abs/2602.15763 After the launch of GLM-5, we’re pulling back the curtain on how it was built. Key innovations include: - DSA Adoption: Significantly reduces training and inference costs while preserving long-context fidelity - Asynchronous RL Infrastructure: Drastically improves post-training efficiency by decoupling generation from training - Agent RL Algorithms: Enables the model to learn from complex, long-horizon interactions more effectively Through these innovations, GLM-5 achieves SOTA performance among open-source models, with particularly strong results in real-world software engineering tasks.

English
0
1
9
2.1K
Jimmy Lin
Jimmy Lin@lintool·
Congratulations to @jietang @ZixuanLi_ and the entire @Zai_org team on the GLM 5 release: based on >6K votes, it’s the best open-weight model on the @yupp_ai leaderboard (with speed control)!
Z.ai@Zai_org

Introducing GLM-5: From Vibe Coding to Agentic Engineering GLM-5 is built for complex systems engineering and long-horizon agentic tasks. Compared to GLM-4.5, it scales from 355B params (32B active) to 744B (40B active), with pre-training data growing from 23T to 28.5T tokens. Try it now: chat.z.ai Weights: huggingface.co/zai-org/GLM-5 Tech Blog: z.ai/blog/glm-5 OpenRouter (Previously Pony Alpha): openrouter.ai/z-ai/glm-5 Rolling out from Coding Plan Max users: z.ai/subscribe

English
8
11
78
22.4K
Jimmy Lin retweetledi
Yupp
Yupp@yupp_ai·
📢 New Model Drop: GLM 5 is now live on Yupp! We've been hosting a cloaked version of this powerful new AI, and it's showed up strong on our user-preference leaderboards – with ~6K votes, it is currently ranking #10 in Text models (with speed control filter on) 📊 Big congrats to the @Zai_org team!
Yupp tweet media
English
16
14
143
19.6K

Keşfet

@zijian42chen @xueguang_ma @ShengyaoZhuang @srchvrs @MiniMax_AI @simonw @yupp_ai @karpathy