theptip

8 ms·

181.
by theptip 5mo ago
Sorry, no. Claude Code was the product that brought coding agents to the mainstream. Sonnet 3.5 was the model and harness that created vibe coding. This was a push of the technical frontier, not a marketing achievement.
182.
by theptip 5mo ago
Honestly I have no idea how you couldn’t tell. Reading a PR I can see the difference without even reading the words. (I doubt I could spot the difference just looking at the code diffs though.) Claude commit messages - well structured test
183.
by theptip 5mo ago
You make a good point, but I think “lowest risk” is also not the metric either. You could buy 100% bonds and be even safer. You want some risk-adjusted return metric parameterized by average returns and also volatility.
184.
by theptip 5mo ago
How is Keybase doing these days? I stopped following after the acquisition, but I like the concept.
185.
by theptip 5mo ago
There are tens if not hundreds of thousands of talented engineers that just got laid off from BigTech. Meanwhile it’s never been easier to found your own company, and if you label it as “AI for X” it’s easy to raise too. From what I see in
186.
by theptip 5mo ago
My read - 4.7 was a tactical lobotomy to improve the average experience at the expense of peak performance; necessary due to compute pressure. Now that they have Colossus capacity, I guess they can tune up the intelligence again and spend m
187.
by theptip 5mo ago
Another (IMO fatal) error is they don’t attempt to measure within-model variance. The thing you find when you actually wire up a rigorous eval is that with tool calls like web search you are wide open to infra issues, flakes, and all sorts
188.
by theptip 5mo ago
I think it’s pretty naive to ask your employer for a day off. As experienced by companies, the market is more competitive than ever. Whoever slows down will get eaten by some hungry upstart that is willing to work 996 to eat the incumbent’s
189.
by theptip 5mo ago
My thought too. This would be huge if true.
190.
by theptip 5mo ago
I agree that providers are in some sense incentivized to juice the numbers here. But, they are in an incredibly competitive 3-way knife fight, and so they are also heavily incentivized to be honest with themselves about quality gaps. I thin
191.
by theptip 5mo ago
> % of AI suggestions accepted vs. edited is also a BS metric I disagree. It’s a valuable metric if you are building an agent / skill infra layer. Think of it like error rate on your API. Green metric does not mean your system is he
192.
by theptip 5mo ago
I still find it lazy and confused vs 4.6. I don’t like adaptive reasoning.
193.
by theptip 5mo ago
To understand the token count thing - spending tokens is necessary and not sufficient to demonstrate that you are adopting AI. Where we were 6mo ago is that a lot of big orgs realized they were behind, and needed some way of measuring if th
194.
by theptip 5mo ago
> Shortening patch cycles will only help so much. It's funny that whenever an NPM supply chain attack is published, people recommend a cooldown before installing new versions, and then when a vulnerability is discovered, everybody j
195.
by theptip 5mo ago
We collectively have to re-learn what operations are expensive and what are cheap. Prototypes are practically free now. You can ask the AI try each architectural or stylistic option and just see which code you like better. To your point, an
196.
by theptip 5mo ago
If the artist sells them directly at the market price set by the resale auction, they get more than if they sell at face value.
197.
by theptip 5mo ago
I think a better model is likely building a full dummy repo and operating within that. It’s now trivial to extract a toy app that is a dummy version of your product system, introduce a sprinkle of bugs, write ok documentation etc - and then
198.
by theptip 5mo ago
I think you should do a bit of research and come back once you have fixed your knowledge gaps.
199.
by theptip 5mo ago
Plenty of analysis on this point, the inputs are not secret. Check out Semianalysis for example. They have something like 70% margin on inference.
200.
by theptip 5mo ago
As always, free cash flow and unit economics are more interesting than net profit. By which metrics Anthropic is making a lot of money.
201.
by theptip 5mo ago
Obviously the pitchforks come out before then. But! Does Musk have his robot army or not at this time? That will determine how the pitchfork mob turns out.
202.
by theptip 5mo ago
They are doing this because they blockaded Iran.
203.
by theptip 5mo ago
I think the idea here is you give the Hunters (stage 2) a narrower scope, but have a parent agent responsible for dividing up the full search space (stage 1). And note that Hunt tasks can be queued from previous Trace tasks, ie you find a v
204.
by theptip 5mo ago
This is cool. I think it fits the Agentic model a lot better; wrangling PATs and repo permissions on GitHub is quite tedious. How are folks wiring up CI/CD? Seems you want to trigger compute DAGs on the patches you receive, what’s the
205.
by theptip 5mo ago
Right, and thence back to the point of TFA. Either you black-box the curve and assume that you will keep stacking sigmoids for about as long as you already have already seen. Or you white box it and make some actual technical argument about
206.
by theptip 5mo ago
Yes, and if you are building a payment API you need to be robust to client bugs.
207.
by theptip 5mo ago
Some sort of eval. Eg TermBench, implemented in Harbor. It’s an insane amount of effort to build shareable, reusable, comprehensive evals, hence why so almost all skills are stuck in the “vibes” phase. That said I think it’s quite easy to s
208.
by theptip 5mo ago
> Rust is a complex language prone to refactoring avalanches This may be so, but LLMs are great at slogging through such tedious repercussions. I would say if the language prevents sloppy intermediate states, that actually makes it more
209.
by theptip 5mo ago
> what should we do with this kind of content? Had the result been produced by a human mathematician, it would definitely have been publishable, so I think it would be wrong to describe it as AI slop. On the other hand, it seems pointles
210.
by theptip 5mo ago
Well, if you actually build the feedback loop then you solve this by weeding out the bad ideas. This is like going to a startup as a senior bigtech engineer - if you can’t ship it becomes clear quite quickly. Regarding the talent dysmorphia

More ›