On openrouter I had to filter out glm 5.3 flash instances running fp4 - making sure it only ran fp8 - as the quantized models kept going crazy / off track.
Also depends which glm 5.3 you use. I used open router and sometimes had good, then appalling performance. It turned out the provider being routed to - some were fp8 and some were fp4. The fp4 ones glitched on anything complex. It is such a difference that I only allow fp8 providers in my allow list.
Because having everything there designed to kill you (even the weeds are deadly, see gympie-gympie) and it being located 3km from the surface of the sun isn't enough, let's factor in externalities as well.
If I wanted to hide out from an apocalypse, I'd move to Heard Island, as long as you're not bothered by the 10% tariffs.
Biggest:
1)stop companies buying up properties and renting them out.
2) big tax on capital gains for any house flip under 2 years holding
3) big companies may own houses in new build projects, with the intention of selling them
4) lean in hard to minenergy prefabricated houses
5) reduce red tape from government
That's it, increase supply, reduce big business demand
They will need to do something new here, somewhere between patents and copyright, to help protect US intellectual property on the AI stage.
Otherwise, US r&d will fold on this.
Longer term, 2 years plus, it will be a race to the cheapest as almost all models will be better than our current frontier ones.
American AI companies pirated hundreds of thousands of books, scraped the web against the express wish of site owners, extracted all public GitHub repositories without the explicit consent of code authors and distilled all of that information, basically the entire knowledge humanity built over thousands of years, into large language models. The way these models were built was ethically questionable. The models are out there now, and time can't be rolled back, but I still don't think the companies who built the models by ignoring intellectual property rights deserve any special protection of their own intellectual property.
How would that be justified, because as I see it would necessarily treat model outputs as more protected than actual human output?
Surely Anthropic/OpenAI ToS isn't something which can have more weight than, let's say, the ToS of a website expressed through robots.txt, considering that for the length, actual human effort has gone into writing the text, at a cost much greater than anything output by an LLM. So so long as the AI firms are allowed to crawl and then train on human-generated websites that don't allow crawling, but are crawled anyway, I think this kind of thing is really hard to justify.
Assume bonsai has the same ticks
reply