theptip

6 ms·

121.
by theptip 3mo ago
Wat. Price/perf has been going down massively over the last few years.
122.
by theptip 3mo ago
Are we reading the same chart? They have Sonnet <= high as Pareto dominant on $/perf. You have to test each task obviously but it is not a bad model on its face.
123.
by theptip 3mo ago
Yeah. It’s not the end of the world. But, it is a big own goal, because once you invest in building evals for your internal use-case, 1) it’s easier to switch your model to whatever is cheapest, and 2) it’s way easier to fine-tune an oss mo
124.
by theptip 3mo ago
I guess I can see why they might nerf detected clients server side, but without evidence I would not assume it. Could also be so that 1) they can identify sus client IPs, 2) do a statistical analysis on distilled models to prove that their
125.
by theptip 4mo ago
> My whole point is that I don't want it to build an entire feature from one prompt You are free to do you. But you were asking about why others want the best model. The answer is, clearly, agentic coding (ie multiple agents each cr
126.
by theptip 4mo ago
But… what effort level? “Opus 4.8” is a massive capability range. If you just ran it on medium that is a completely different result than vs. max.
127.
by theptip 4mo ago
This seems to be the crux, I couldn’t find a place where the bill explicitly says, so AIUI the rule making could fall either way. Are you “compensated for being oncall” if your contract says you may do some unspecified amount of oncall, and
128.
by theptip 4mo ago
This seems mostly good for restaurants, some concerns I had from the title seem to be handled reasonably. It’s not preventing “can anyone cover Saturday” messages in a group chat. Just the case where shift changes are made and workers are _
129.
by theptip 4mo ago
I for sure agree that plenty of current use-cases are solvable by non-frontier models. However, you said “new versions with features that nobody asked for”, and I would prefer that you concede the point before shifting to arguing a new poin
130.
by theptip 4mo ago
Eval saturation. “Alice is supposedly smarter than Bob, but they both take the same time to tie their shoes.”
131.
by theptip 4mo ago
But in this model, there is no regulator, there is just the Whitehouse deciding who gets to use the AI. Nobody has “captured” Trump here.
132.
by theptip 4mo ago
“More intelligence” is the new feature. Almost everyone is asking for this. Citation: have you looked at OAI and Anthropic’s customer growth numbers?
133.
by theptip 4mo ago
No. As long as you downloaded the weights, you can run them somewhere.
134.
by theptip 4mo ago
> Maybe it’s the realization that it was never that cheap in the first place and they're forcing us to upgrade in a slow and painful way. All the analysis I have seen points to frontier models being profitable to serve. It’s using 5
135.
by theptip 4mo ago
You have wind right?
136.
by theptip 4mo ago
Why do you feel the important part _now_ is where the weights get run? I can see this as a future battleground but access to frontier models (which you cannot run locally) seems a lot more relevant today.
137.
by theptip 4mo ago
> I haven't seen anyone say the best education would be pure-human provided I took GP to be saying exactly that, and the OP’s full ban on AI in schools is also enforcing pure-human. > I don't know if you used AI to learn abo
138.
by theptip 4mo ago
I agree it’s an option. I did some essay exams at Uni too. I just don’t think we should assume that the old cost/benefit calculus still applies and this is the best way to test e.g. English Lit or Philosophy.
139.
by theptip 4mo ago
That’s our job as educators, not the kid’s. “The correct way” obviously requires scaffolding.
140.
by theptip 4mo ago
I agree that we are seeing the negative effects of AI-done-wrong. (The best technology ever invented for avoiding learning, as I said.) I don’t agree that this means all potential or current uses of AI are harmful. I also don’t agree that s
141.
by theptip 4mo ago
I agree that the human touch is valuable. I disagree that humans are required to bootstrap meta-learning; I think it is quite deeply wired into the human brain and there is no reason that we can’t create digital gyms that give opportunities
142.
by theptip 4mo ago
I thought the dichotomy would make this clear, but the claim was not that however you use AI, it’s the best thing ever. To be more explicit: _used correctly_ and to the fullest of its capabilities, AI is the best technology ever invented fo
143.
by theptip 4mo ago
But we do give them bicycles, so the lesson would be “give them technology that is safe to use”. And then we are back to the discussion at hand none the wiser; what forms of AI are safe to use for kids?
144.
by theptip 4mo ago
It’s a good point. I could see a world where exams get cheaper to administer. I also think we can use project work as an ongoing assessment.
145.
by theptip 4mo ago
It’s one dimension, I agree. Though, for essay-based courses I’m a little skeptical about writing under time pressure. My thinking here is that you might want to also introduce more “viva voce” (verbal defense) style tests. These have been
146.
by theptip 4mo ago
> Generative AI is not going to help them with those skills. I think it's more complex than this. AI is both the best technology ever invented for avoiding learning, and the best technology ever invented for learning. The cat is out
147.
by theptip 4mo ago
> If you show revenue, people will ask 'HOW MUCH?' and it will never be enough. The company that was the 100xer, the 1000xer is suddenly the 2x dog. But if you have NO revenue, you can say you're pre-revenue! You're a
148.
by theptip 4mo ago
Yeah, plenty of rigorous work out there. https://transformer-circuits.pub/ is the OG. A good recent-ish paper was https://www.anthropic.com/research/alignment-faking . But my comments about generalizati
149.
by theptip 4mo ago
Right. The fact that some good stuff is everywhere because of social media campaigns, doesn’t mean all the good stuff is everywhere. But, for many the base level of content recommended is good enough and so they are not forced to hunt. So t
150.
by theptip 4mo ago
> a single type would have two projections: a value variant (flat, never null, behaving like a primitive) and a reference variant (a box that allows null). Across various iterations this was written as Point.val/Point.ref, and later

More ›