(in 2015) scraped data from around the web, converted it into clean, real-time JSON feeds, then sold those to analysts and media cos Was a nifty business - though scaling/maintenance was hard in the pre LLM days

*sniff* *pulls shirt* You know, this is perfect - *gestures wildly* - this is the ultimate perversity of capitalism at its purest. Here we have Anthropic, this company claiming to build "AI for humanity," and what do they do? They create this digital cocaine, this Claude Code,

This has been me post Opus degradation It’s weird because even a degraded Opus is better than the alternatives rn. But tasting what a great AI programming agent could be like, and having that taken away - that’s been bitter Hoping for a good-at-agentic-work GPT5 soon 🤞

The new OAI models significantly exceed my expectations - very little yapping when thinking not needed. much better then qwen here ("user confused. need clarify") - lightning fast (40 tokens/sec on my M2 Max while on low battery mode) - extremely good at SQL and code

Rishabh Srivastava Since you mention fine-tuning, we optimised that path a bit (loads of coo kernel magic happening behind the scene): cookbook.openai.com/articles/gpt-o…

Sheesh, they weren't kidding about GPT-5 being better at tool use. Have not seen an openai model make this many tool calls yet Also seems really good at making parallel tool calls compared to Opus (atleast from early vibes) Will check with Codex soon

Early vibes: Codex CLI (w GPT-5) can now do tasks of similar complexity as Claude Code (w Opus). CC has way better UX though Excited to try out OpenHands and AmpCode with GPT5. More competition FTW!

GPT5 works great for me via the API (explicitly setting reasoning levels / verbosity), but seems like a regression over o3 on ChatGPT :/

Now that the era of the scaling "law" is coming to a close, I guess every lab will have their Llama 4 moment. Grok had theirs. OpenAI just had theirs too.

Jfc what is going on at OpenAI. Ran Codex in "ask me for approval for every edit" mode - got this kind of impossible to read text back, every time I did. Seems like something broke in the last couple of days - no idea how tf everything is so broken.

new blog! i got mad at openai and vibe-coded a triton kernel

Credit where it's due: seems like OpenAI has fixed a lot of GPT-5 issues in the last 12-24 hours, and Codex CLI works really well in auto mode Still terrible if you use in a "approve before making edits" mode, but hopefully they fix it soon🤞🏼

The most interesting tension in the AI industry: - better lesson pushes models towards end to end solutions with single large models - economic viability pushes industry towards many composable models with broad horizontal applicability

One reason I'm very bullish on Cloudflare Workers and Durable Objects - solves this in a very meaningful way