p1esk

12 ms·

271.
by p1esk 1y ago
It depends on how many kids his descendants had. If it were, say, five kids on average, and we are tracing an average descendant, then the age of conception of child #3 can well be 33 years old.
272.
by p1esk 1y ago
What was the average age of conception for the last child?
273.
by p1esk 1y ago
Did making things without machinery require fewer skilled people?
274.
by p1esk 1y ago
Prompt injection is “getting it to ignore instructions”. You’re contradicting yourself.
275.
by p1esk 1y ago
One step closer to mass unemployment, yay!
276.
by p1esk 1y ago
If I eat raw salmon (sashimi) a couple times a week, would I still benefit from consuming fish oil?
277.
by p1esk 1y ago
10 years ago I used LSTMs for music generation. Worked pretty well for short MIDI snippets (30-60 seconds).
278.
by p1esk 1y ago
I’m not “elite” of any kind.
279.
by p1esk 1y ago
We are the elites Wait, what?
280.
Canaries in the Coal Mine? Recent Employment Effects of AI [pdf](digitaleconomy.stanford.edu)
102 points by p1esk 1y ago|110 comments
281.
by p1esk 1y ago
Why was it bad?
282.
by p1esk 1y ago
“Designed by AI”
283.
by p1esk 1y ago
I take a shower at the beach, bike back (~5 min), change, and start checking my email 5 minutes later (I work remotely).
284.
by p1esk 1y ago
They experimented with gpt-2 scale models. Hard to make any meaningful conclusions in the gpt-5 era.
285.
by p1esk 1y ago
What would you expect?
286.
by p1esk 1y ago
I wake up at 7am, take kids to school, go to a gym 8-9, go for a swim in the ocean 9:15-9:45, start working at 10. Feel great.
287.
by p1esk 1y ago
GPUs completely dominate the cost of a server, so a GPU upgrade typically means new servers.
288.
by p1esk 1y ago
How do you know?
289.
by p1esk 1y ago
Yes, I’d be curious about his experience with GPT-5 Thinking model. So far I haven’t seen any blunders from it.
290.
by p1esk 1y ago
In the world where you do not claim that LLMs suck today based on your attempt to use some shitty model three years ago.
291.
by p1esk 1y ago
With good quantization you can get 36GB down to 8GB. To get 36B down to 8B you need good pruning.
292.
by p1esk 1y ago
Did you use one of the latest frontier reasoning models? If not, how is your experience relevant?
293.
by p1esk 1y ago
Wait, you decided to pay $200 a month and you didn’t know that o3 was better than 4o?
294.
by p1esk 1y ago
Are you saying they have already trained gpt-5 on the entirety of world’s video data?
295.
by p1esk 1y ago
He didn’t describe any fundamental issues.
296.
by p1esk 1y ago
I don't think anything with mechanical moving parts is going to last that long, with regular usage, and have original parts. You should visit any third world country: plenty of old cars still running around.
297.
by p1esk 1y ago
Absolutely. There are tons of young talented players who just need room to shine
298.
by p1esk 1y ago
I don’t actually, did he specify a year? He made a lot of predictions, many are wrong, but the AGI one is pretty amazing.
299.
by p1esk 1y ago
You can easily hire 250 people with doctorates in LLMs if you’re willing to pay $1M salaries. But they are not going to be like this guy.
300.
by p1esk 1y ago
Ilya was offered 32B

More ›