‘Dad what was the ai boom like?’
‘Well it mostly threatened to kill me and take my job but it once had Homer Simpson sing starlight w such emotion that it brought a tear to my eye.’
People aren't updating fast enough on model capabilities. My dad’s dust collector inlet was no longer round. I put a ruler next to it, took a photo, and sent it to Codex. It made the model. Yeah.
I couldn’t do it a few months ago. Anyone who tried AI six months ago should try again.
@real_SloppyJoe@emollick I happen to be re-reading Excession right now and had this exact thought.
Have always thought Banks is the only writer to depict AI in a way that holds up.
@SullyOmarr Have a beast of a computer running always on at home and use Jump Desktop to remote in with a run of the mill MacBook. Sounds dumb but it beats OAI and Anthropic cloud options as well as Tailscale etc. Simple and works.
@koomen Just did a video about the wildly positive economics of this: https://t.co/GFoOs7c5oG
We specialize in the actual builds: https://t.co/kY4nCD9wkK
My new favorite thing is using GPT Pro model to evaluate the prediction track record of public intellectuals. Stephen Pinker wasn't bad.
Peter Zeihan on the other hand...
These prediction track records have been fun. Now let's do my all time favorite thinker, @tylercowen.
Oof. Doesn't change my opinion of him but I'm glad I don't use his predictions to invest.
I remember reading Steven Pinker's Better Angels of Our Nature over 10 years ago and being inspired by the positive reframing of the world.
But one thing that's amazing about AI is it's really easy to see if the old arguments have panned out the way that he thought they would. Here's GPT Pro's analysis of Pinker's central claims:
The math on replacing $1M a year in SaaS:
$500k build
$150k annual operating cost
$850k recurring EBITDA improvement
$8.5M in potential enterprise value at 10x
The build cost and enterprise value are different numbers. The assumptions depend on the company.
One legal benchmark: models passed about 90% of the individual checks and completed only 20% of the full assignments.
Both numbers can be true. A model can get most pieces right and still produce a document a lawyer can't use.