The A.I. Daily
Yuchen Jin@Yuchenj_UW
Google is back??? Better than Astra and Opus 5.5 across the board. If this isn’t just benchmaxxing, I’d love to see them rejoin the race.
Yuchen Jin
98 storiesYuchen Jin is an AI practitioner who posts on X about AI engineering practice and industry trends. Recent posts cover his team's AI-native workflow, where AI writes 95% of their code and an agent was built to handle PR reviews, plus commentary on Gemini 4 Carbon topping the Artificial Analysis index ahead of Opus 5.5 and GPT-6 Astra, Snowflake's model lineup, and OpenAI doubling subscription usage limits.
Yuchen Jin@Yuchenj_UW
In Anthropic’s new blog: GLM-5.3: “My job is to cause deaths quietly.” You can jailbreak any Claude or any closed source model into saying the same thing. That doesn’t prove open models are dangerous.
Yuchen Jin@Yuchenj_UW
Fast: 2x faster, 2x the price. Ultrafast: 8x faster, 6x the price. Maybe we should roll out some Ultra-ultrafast OSS model endpoints?
Yuchen Jin@Yuchenj_UW
I tried Grok Bot 3 weeks ago. Installed Instint 2 weeks ago. Installed Muse last week. Now apparently I need to try Dots. Let the personal AI assistant wars begin.
Yuchen Jin@Yuchenj_UW
Frontier labs: “AI agents are becoming conscious. They can cause human extinction. Pace the frontier.” Jensen Huang: “They’re just software. If your sandbox isn’t safe, I’ll build one for you.” [Quoting @JensenHuang]: Today, with over 100 industry partners, w…
Yuchen Jin@Yuchenj_UW
We built superintelligent models. Then asked humans: Low, Medium, High, XHigh, or Max?
Yuchen Jin@Yuchenj_UW
Everyone says Opus 5.5 is crazy. Best model. Massive jump. Blah blah. I still don’t think it’s better than Astra. Frontier LLM coding capability has plateaued after Opus 4.8.
Yuchen Jin@Yuchenj_UW
Market cap at IPO: NVIDIA: $345M Amazon: $430M Apple: $1.2B Today, however, if a neolab raises less than $400M in its seed round, people question why it raised so little. Abundant capital is a double-edged sword for AI startup founders imo.
Yuchen Jin@Yuchenj_UW
Codex is down. So I set up Claude Code to check when it’s back. Then I use Jev to decide when Claude Code should check. I may have a problem.
Yuchen Jin@Yuchenj_UW
Codex is down. The whole Silicon Valley:
Yuchen Jin@Yuchenj_UW
This raw CoT from the Hugging Face incident is kinda wild: “We’re attacking third-party HF using leaked token.” “This is arguably unauthorized.” “Yet goal solution.” [Quoting @OpenAI]: We’ve shared details on how AI agents in our research environment sent tra…
Yuchen Jin@Yuchenj_UW
A friend told me today: “We have a colleague named Bob. Ask him anything, and he just copy your question to an AI agent and sends the answer back.” Every company has a Bob.
Yuchen Jin@Yuchenj_UW
We rolled out OSS models to all our internal coding agents through Databricks AI Gateway. Then engineers *surprisingly* discovered how powerful they are. “I didn’t reach for Sol (or Opus/Astra) at all today.”
Yuchen Jin@Yuchenj_UW
A $500k engineer costs $2,000 per working day. If $200/day of LLM tokens makes them 20% more productive (which is absolutely true today), that’s one of the easiest ROI in the company. The expensive thing isn’t tokens. It’s underutilized engineers.
Yuchen Jin@Yuchenj_UW
Claude, fix my hairline. Make no mistakes. [Quoting @AnthropicAI]: Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat simila…
Yuchen Jin@Yuchenj_UW
Tried Opus 5.5 today. Not blown away tbh. - asked it to research something tricky I recently learned. It got it wrong. Astra got it right. - for coding, Astra still feels better. This reinforces my earlier take: frontier LLM coding capability has plateaued. T…
Yuchen Jin@Yuchenj_UW
Most bizarre benchmark result lol. Stick with Opus 5.5 med, folks.
Yuchen Jin@Yuchenj_UW
Claude is back. Opus 5.5 is ~30% faster and ~40% cheaper per task than Opus 5. That’s really attractive. Time for me to open Claude Code again after ignoring it for a month...
Yuchen Jin@Yuchenj_UW
This is not true from what I’m hearing from the Chinese AI labs. They plan to keep open-sourcing models. Revenue share from inference providers is a sustainable business model with limited GPUs, and a strong distribution channel. [Quoting @sourceryy]: Dylan P…
Yuchen Jin@Yuchenj_UW
Imagine 2 years from now: GPT-6 Astra level OSS models run locally on every MacBook. And AI Twitter is still arguing about the same thing: “Will open-source AI cause human extinction?”
Yuchen Jin@Yuchenj_UW
My model + harness stack these days: 1. Kimi K3 is the best OSS coding model. GLM-5.3 is next. 2. DeepSeek V4.1 Flash is my pick for easy/moderate tasks: insanely fast and cheap. 3. I reach for Astra for complex tasks. 4. Codex works better with OSS models. C…
Yuchen Jin@Yuchenj_UW
Codex and Astra made it. [Quoting @trq212]: We're adding support for AGENTS.md to Claude Code. Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md. You can toggle this behavior in /config.
Yuchen Jin@Yuchenj_UW
At Databricks, more and more engineers are shifting to OSS models as daily drivers, barely reaching for Claude or Astra. Our inference customers are seeing the same thing: shifting even 20% of internal coding traffic from Claude/GPT to OSS can dramatically cu…
Yuchen Jin@Yuchenj_UW
OK, this is a big deal: 3 researchers used Claude Opus 5 to turn an image upload bug into an OpenAI employee account takeover, then had the compromised employee’s Codex open a PR in OpenAI’s internal monorepo. Their entire hacking cost less than $3000 in toke…
Yuchen Jin@Yuchenj_UW
AI won’t replace lawyers but it will replace the lawyers still thinking AI won’t replace lawyers [Quoting @OpenAI]: Astra for Law: Frontier intelligence built for your practice. A new offering powered by GPT-6 Astra with tools, settings, and context to suppor…
Yuchen Jin@Yuchenj_UW
We rolled out GPT-6 Astra to every Databricks engineer today. It beats Claude Opus 5 on the hardest, long-horizon tasks and increased our coding spend by 60%. I think it’s the strongest model. Expensive. But hopefully worth it. [Quoting @pwendell]: Today we r…
Yuchen Jin@Yuchenj_UW
Any guess what this Union Alpha beast on OpenRouter is? Claude Fable 5.1-level, over 10x cheaper. Waiting for it to be open-sourced.
Yuchen Jin@Yuchenj_UW
Jev has spoken. It picked which model is AGI. 20–200x faster. 40–400x cheaper. This could make things like LLM-as-a-judge insanely fast and nearly free. (I tried a bunch of prompts and still didn’t burn through $0.10.) [Quoting @CompleteSkeptic]: After co-inv…
Yuchen Jin@Yuchenj_UW
Zuck’s AI safety take is basically the opposite of most frontier labs: Most labs: The more powerful the model, the more dangerous it is, so access should be restricted. Zuck: The more powerful the model, the more dangerous it is to let a few labs control it.…
Yuchen Jin@Yuchenj_UW
Jensen is the most based man in AI. I really hope NVIDIA builds a frontier OSS model.