AGI before GTA 6

Don't do this. One of my gripes of modern web programming is the lack of fundamentals. Use HTTP!! Don't be overly verbose, why success? Why error? Why waste that precious bandwidth and memory on unnecessary fluff.

The Chinese way of promoting bureaucrats with tests might be the best. Mayors+ should have at least a college degree and some training in leading a big city. We should however NOT create a universal test or enforce single party rule. Something in between; reward talent,

Peer Richelsen — oss/acc climate friendly unfortunately rebranded to degrowth so it's completely done in the U.S. at least. Europe's climate policies have reduced it to a joke that lets 100k+ people die of heat rather than install ACs.

i think Scholarly will eventually merge all the existing features into a single website creator for users to use. I explored this a while back, but models were just not there yet. They required too much feedback, more tokens, more $$. Not profitable for a $20/mo plan.

Gemini will be the truly next generation model. The model generations are: GPT 3.5, GPT 4.0 and o1/o3. o3 was an extension of o1, more RL. If gemini 3.0 is not a major cheap, ~20% increase in Arc AGI // HLE then we're at least ~5 years out from AGI.

We'll probably have real-time image enhancement within 1-2 years. Old games will be remastered via an AI filter. It'll be incredible.

(:

My Samsung OLED monitor is exactly like this. Horrible UX, randomly turns off, etc. Only reason I have it is so I can switch between Windows desktop & Mac.

Oh, DMS have gotten a facelift. It's actually so much better now. Nikita Bier any plans to bring Grok to chime into DMS, especially group dms?? 👀

Perplexity's ads are just depressing. Comet browser still has it's quirks, it's not ready yet. They're trying to get as many users in and hope they stick with a subpar product. i dont get it.

Steam Machine…

Seems quite interesting: They're probably using diffusion to generate gaussian splats. I'm not familiar w/ gaussian splats so there might be an intermediatory step in between.

Politics at Apple seems to be at the tipping point. Someone obviously put a fake story to the WSJ in order to push Tim Cook out. Tim has been an amazing operator, but people within the company want him out. 👀

I wonder which lab will release the first multi-modal reasoning model. I've read a few papers, but so far there's been no real scaling up of multi modal reasoning.

Optimizing LLMs is such an interesting process. Simply adding a word to the system prompt will decrease costs by up to 60%. "LLM Costs Optimizer" will be a role in the future.

Dealing with EU customers is one of the biggest pains. Eventually, most startups will probably just block EU users. It’s just not worth the time or hassle.