Trunk Monkey

This may be the coolest emergent capability I've seen in a video model. Veo 3 can take a series of text instructions added to an image frame, understand them, and execute in sequence. Prompt was "immediately delete instructions in white on the first frame and execute in order"

Let's compare Qwen 3 Coder & Sonnet 4 for code generation:

Silly debate. Now if they would have said vs 100 Scots or Irish, that would be a conversation.

Why does @SlackHQ search have to be like it is? It doesn't give me the same context multiple times in a row even if I pick a particular channel. There is a zoom call link for instance that I have to search for twice every time I need to use it because the first time forces…

Even in the early 2000's most BIOS code was crazy spaghetti legacy crap. Big reason that most of the dev work went to Asia is because it was impossible to hire for. All the original dudes got rich off stock options or moved to senior roles and nobody could replace them.

I want less radius, not more. Why do I want to see my background on full screen windows?

If they cant handle that launch, they shouldnt be out in the ocean.

Dude, new startup idea! AI wedding DJ. Put in you and your families personal deets, music genre, activities and time schedule. Have it listen to the vibe and adjust. Intro the speakers. Cut the mic on drunk uncle. Just cut me in for a few shares 😆

Meta prompting deepseek-r1:8b-0528-qwen3-fp16 gave me one of the most hilarious self doubting, off the rails sessions I have seen in a long time. Over 50k tokens of this..

I used public wifi without a VPN and a hacker SQL injected me with coronavirus

Isn't that the generation with the tech troubleshooting skills on the same level as boomers? They can just pipe down, eh.

Also this album makes zero sense if the cornucopia never existed. discogs.com/release/372562…

Yep as an extremely heavy user, Claude is still the best for some things but you have to but in a ton of work and tooling from other models to get it there.

Sonnet has been so far ahead that I haven't used any of the openai models for a while. I have branched out to Gemini 2.5 pro and grok but mostly use those for planners to keep sonnet on the rails. If you really want to know the best combo, aider polyglot bench seems most…

You know what is also really good but people aren't talking about? Grok. I like grok better than sonnet as a planner (Gemini is still champ here) but equal to sonnet as a coder. I will switch back and forth sometimes if a model is stuck or in a loop and both are very equal.

If anyone is looking, it's incorrect on the ollama site, this is what you need: ollama run deepcoder:14b

Have they done a stealth update with Sonnet 3.7? The past few days, it has become incredibly lazy. Removing or mocking code errors, constantly creating new filenames while fixing issues in prod code, quickly giving up and declaring victory when encountering simple errors. This…

Noted some issues with Gemini 2.5 Pro not digging into some large context tests I threw at it on launch day. It was pretty bad. Been pushing it all day today though and very impressed. Seems to be a bit better than grok at finding issues or planning features in a large codebase.

A Story in 3 parts. OpenAI x.com/Josikinz/statu… Claude x.com/Josikinz/statu… Grok x.com/Josikinz/statu…

I did a side by side test from the UI of grok and Gemini pro of a RepoPrompt compiled codebase of a little less than a million tokens. Used a simple 'find bugs' prompt. Grok thinking gave me an extremely detailed report of 7 legitimate issues it found, details of how to fix…