Excited for grok3 to be added to Cursor has become my go to for debugging. Seems much more adept than even sonnet 3.7 for understand overall context of files, maybe due to larger context window? Only annoying thing is not having it in composer...

Spacex: Catches a massive supersonic object. Me: "No sonnet, that didn't fix the type issue please try again"

vibe coding is fun until theres a bug the models cant solve and you have to understand the codebase.

I wish I had a blood pressure cuff readily available so I could see what my BP is after spending hours diagnosing a lambda - elasticache VPC connection issue...

Is gpt 4o image really that much better than gemini 2.0, midjourney, or any of the others? Or is it simply that people weren't using those for images and now they can get images?

Coding with cursor is 90% efficiency increasing. The other 10% is when I am waiting for an answer to a question so I leave the IDE and end up getting pulled away.

Lock in, fellas. The B2B SaaS market just got a whole lot bigger.

finally Grok is in cursor! Api seems more stable than the platform too...

when people talk about sonnet coding for hours, I just can't imagine the labyrinth of debugging waiting when something breaks in 6 months.

Running my first claude code PR. Wish me luck

Gm excited to put my Claude code devs to work today

this is a great idea. sometimes I just put in random destinations when I want to drive around and think. Similar to going for a walk, mindless but changing scenery causes ideas.

Come hang out with us & Model Y for 30 mins Full drive in 1x speed below

Gemini models are really awesome and fast, but the pay as you go random rate limiting makes them almost unusable from a cost perspective until you are running tens of billions of tokens per month. Unfortunate.

even gpt-5-agi can't fix CORS errors

testing gemma 270M reminds me how dumb LLMs used to be...

x.com/i/article/1965…

Rather than hoping new models have higher throughput I am now looking to get better performance with less context. Used to push context, but I think better agents will start being more selective, not quite RAG but something in between...

The tricky thing with context is that, you need to put everything in whereas humans see much more context than we read. Imagine looking for information on a website. Probably 20-50k tokens of "context" but you skip 95%+ of that. IMO this is the current shortcomings of most agents

go to try out new OpenAI connectors with spotify, paste in 19 songs someone on reddited listed and asked to make a playlist from them (they are all well labeled)... guess we're not quite there yet