jmoiron plays the blues

@jmoiron.bsky.social replied: OTOH, if inference is already a bad business how are bigger models going to make it better? Not that it's surprising for delusional SV narcissists to play roulette with the entire US economy, but if this is true and it all blows up, I don't see what distinguishes it from eg. FTX.
@jmoiron.bsky.social replied: If inference is already super profitable then I don't understand why the practice around quotas is so vague and they've been pulling back access consistently over the past few years. That feels a lot more like a company tuning for profitability/oversubscription.
@jmoiron.bsky.social replied: Unlike "How will AI impact jobs?", the answers to these questions are knowable; there are people who know them already! But without data, their claims are indistinguishable from the kinds of misinformation you get from elusive marketing speak or the overconfident vibes of laypeo...
@jmoiron.bsky.social replied: There are other basic questions that are frustratingly inconclusive, too. I don't buy LLMs as a path to AGI, personally, but have we hit the scaling wall yet? There is a long history of blog posts claiming we have! GPT-5 leads me to suspect we have. But I don't really know.
@jmoiron.bsky.social replied: I don't think everyone would be rushing to build more expensive frontier models if the current inference business model didn't work. But if you listen to their rhetoric, they are insistent on the valuation friendly story that the hockey stick continues far into the future.
@jmoiron.bsky.social I've seen people very confidently claim two opposing narratives on inference: * AI inference is in a VC backed burn phase and the real cost of inference requires pricing people won't pay * AI inference prints money and AI company burn rate is mostly due to training new frontier ...
reposted @bryanl.dev: If coding is your favorite part of software engineering, keep coding. AI doesn't stop you. But if coding was the only part you were good at, that's a different conversation. The job was always bigger than the code. #bransoncognac blog.bryanl.dev/posts/ai-sen...
@jmoiron.bsky.social replied: It's like a war movie in that sense. I was 16 when I came out and I was easily fooled by its rebellious anti-consumerism into viewing project mayhem as a valiant struggle. Fully understanding it in my 20s and reckoning with my failure informed how carefully I navigate political...
@jmoiron.bsky.social I didn't expect "agentic AI" to usher in a resurgence of the Unix command line, but it makes sense in hindsight, as composition has more leverage now. It's also resulted in a rush to poke holes in the walled gardens that have been closed since the death of web2 APIs. github.com...
by jason moiron, 2026