Cherry Plums” are my favorite out of the stone fruit hybrids (Apriums, pluots, etc) https://cdn.nostrcheck.me/cf7c3849ce324fbac4b08c6794a406a8a9abe6fc27e65c62061b3036cd686e47/b31ba9b6ed5c1d2334aa5d2ebac4e828572e07d82aaabf61c40018066c4289a5.webp
In hindsight, the TV shows TechTV, MythBusters, and No Reservations influenced me a lot.
I just need Opus 4 to tide me over until GPT-5.
I just need GPT-5 to tide me over until Gemini 3.
I just need Gemini 3 to tide me over until Opus 5.
Software 1.0 = Coding
Software 2.0 = Model Weights
Software 3.0 = Prompting
Be fluent in all of them.
Andrej Karpathy: Software Is Changing (Again)
https://www.youtube.com/watch?v=LCEmiRjPEtQ https://cdn.nostrcheck.me/cf7c3849ce324fbac4b08c6794a406a8a9abe6fc27e65c62061b3036cd686e47/8e644ed803d943475f524c1cb2a42565120679a81432f3e0466328336e80c01e.webp
They call it “Claude Code” but people are turning it into a general purpose LLM scripting platform.
The main advantage it has over the desktop apps is it can “auto compact” when the messages get long and keep going and going.
What would you do if you had a million tokens?
I'll tell you what I'd do, man: two Claude Code windows at the same time, man.
AI is following a similar trajectory as smartphones
First, smartphones were enthusiastically adopted by power users who got a huge stepwise advantage (blackberry execs)
Then huge skepticism by general public
Followed by utter domination
Finally collective amnesia that the world was any other way
We're on an Intel style Tick–tock model with these AI releases now.
Tick: new OpenAI model
Tock: new Anthropic model
The cost difference between using Claude Code and cursor /API is insane.
For the $100/mo plan I've already gotten $500+ dollars worth of tokens in 3 days.
I’m basically on track to spend ~3k worth of API credits per month. https://cdn.nostrcheck.me/cf7c3849ce324fbac4b08c6794a406a8a9abe6fc27e65c62061b3036cd686e47/9648160592198fdcf2e043b5790e65456d4ac444f9b00c42fadd1901ce35cefc.webp
Balsamic vinegar gets all the attention but I'm more into red wine and champagne vinegar now https://cdn.nostrcheck.me/cf7c3849ce324fbac4b08c6794a406a8a9abe6fc27e65c62061b3036cd686e47/f716afa476c84a1f2a9410eeab3c2627a975e3e71442f39d434ecd2a5b816293.webp https://cdn.nostrcheck.me/cf7c3849ce324fbac4b08c6794a406a8a9abe6fc27e65c62061b3036cd686e47/b2aa2b820e679ab691787194a4f57f4120211602812bb5adfef1e8f8f1c08c31.webp
- Wake up
- Max out Claude Code credits
- Do other stuff
- Max out Claude Code credits
- Do other stuff
- Max out Claude Code credits
- Do other stuff
- Sleep
Opus 4 is about 8x more expensive than o3 now (after the o3 price drop).
The lack of auto advance on short form video platforms is interesting.
In contrast regular YouTube keeps playing on its own forever.
Guessing the manual user swipes are an incredibly powerful signal for the algo.
Does anyone actually want the video to repeat by default?
No matter how well intentioned your company is, some random judge can force you to spy on everyone.
I’ve been vibe coding since the GPT-3 API beta.
The AI cope bubble is getting even bigger than the AI hype bubble.
I feel it’s significant especially when it influences what people eat.
I agree with you on the pharmaceuticals. But there’s lot of good counter research coming out. Follow up studies on SSRIs are very interesting for instance.
It’s sometimes useful for emerging fields where there’s no proper research. For example bodybuilders in the 80s got a ton of stuff right. And it took the research 30 years to catch up. But when good research exists I like to use it.