Josh Dickson

if you think AI chatter has reached an annoying level right now you're in for something else. it's going to be the only thing on anybody's mind starting shortly

You know you’ve made it when every submission to app review is headline news 9to5mac.com/2025/05/09/no-…

There's no substitute for shipping often

Learnt today that you can type any nonsense into Google followed by "meaning", and AI will assume you're searching a well-known human phrase and frantically come up with what it thinks it could mean.

For me, it's the right balance of: 1. Eagerness to write excellent code and follow instructions 2. Ability to explore boundary ideas and options while not getting overly distracted 3. Willingness to challenge my questions or ideas and play devil's advocate

After 10+ hours of use the last couple of days, for my work flow, Gemini 2.5 Pro is that good.

That said, the UI is so confusing in AI Studio I have no idea if I'm being charged to use it and if so what bill. It quietly threw away a conversation without warning (didn't expect that). Long way to go.

Gemini Pro 2.5 is really, really excellent. It's what I was hoping Sonnet would be in 3.7. The right amount of programming help with enough disagreeing, suggestions, and background information. And remains on task and isn't so easily distracted.

Truly the best and worst DX of 2025 so far

Wish I loved anything as much as GPT 4.5 loves an emoji-bulleted list

Sonnet 3.7 "Unexpected capacity constraints" at 12:05AM Eastern Standard is wild

@jorilallo we were promised flying cars, instead we got 5 req/min in US-East

grok-3: no API, yet 3.7-sonnet: rate limits to make it useless gpt4.5: $75/1M tokens 🫠

Productive day (the limit is pretty high!)

On a long enough timeline, everyone sells ads x.com/modestproposal…

OpenAI is going to be an incredible advertising business

Deep Research from OpenAI seems like the most interesting AI product advance in a while. I'm really liking the outputs I'm getting so far. I wonder how much of the advance is about real-time access to the complete text of web pages, and how much is about more extensive CoT.

Very difficult to scale R1 inference at the moment at anywhere close to DeepSeek pricing!

growingdaniel thefp.com/p/exclusive-da…

It’s your turn Anthropic! wallpaper.com/tech/openai-ha…

Finding DeepSeek really helpful to iterate on reasoning prompts that o3-*/o1-* refuse to answer due to (wrongly!) flagged terms-of-use / usage-policies issues.