when you are using vibe coding you are actually removing a very important part of something that is the learning loop this is the bit that worries me the most with too much vibe coding. I think you still need to understand, at least at a high level, how things work

I've just tried Nano Banana 3 and it managed to generate this picture of the Patoruzu comic strip. I’m surprised it recognised the character so well, it kept the classic style and even put together a story in spanish that sort of makes sense

this is a great use case for UI designers, sort of figma but directly in the IDE

most based CEO ever!

I think it’s underrated how insanely good Grok is at explaining posts, it feels like people don’t value it enough but it’s going to be a fundamental part of any information system that uses LLMs to get context right

I didn’t know the Netherlands taxes options when the shares can be sold, which ends up close to exercise. In the UK things are easier under the EMI scheme, gains that qualify for BADR are taxed at 14% (they were 10% until last budget) and I heard they will increase to 18%

I found Grok works well even with posts that have sarcasm. It’s probably easier for 𝕏 since it owns all the data and can build good context The post below for instance , it gave a good answer even though it’s just a dot

I’m wondering if someone’s already working on a peer-to-peer distributed inference. Probably too early but hopefully it exists one day is anyone working on that already?

this must be the reason why they disable it now, I managed to take a screenshot of mine

MCP UI is such a good initiative, there is no standard for doing LLM inference in the frontend other than the event stream format, every provider handles it differently

I’m wondering if this might be linked to the issue I’ve seen in some LLMs where they tend to be a bit too agreeable even when the user suggests bad ideas

I think OLMo is a good model to watch closely. Open source, everything released and already near the top on reasoning and math + fine tuning it on your own data costs a fraction of the closed alternatives

Google has a great opportunity to improve their AI mode search and make it more inline to their search results like 𝕏 does it with the grok button in each post

wow! the bar keeps getting raised!

I really like the demo video from Claude Opus 4.5,in 94 secs it shows how good the model is for: - context window usage - time taken - total cost to run the task pretty fundamental for any new LLM especially around coding which is what they do best great job Anthropic

dogs really see something that we don’t see

an absolute must read for anyone trying to be more productive when doing work

I’ve just discovered what balut is 😑