Hacker Timesnew | past | comments | ask | show | jobs | submit | adriancooney's commentslogin

There are real productivity gains by using these tools right now. Instead of doing 1x your normal work, you can do 5x while still maintaining quality. This is like an accountant sticking to pen and paper because calculators are big and clunky.

In your analogy that calculator would only produce a correct answer 80% of the time, and plausible looking but incorrect ones the other 20%.

If that were the case I’d hire pen guy.


What's the error rate of the pen guy?

Also, if your AI has a 20% error rate, you're not holding it right. You need to spend more time keeping it on rails - unit tests, integration tests, e2e tests, local dev + browser use, preview deployments, staging environments, phased rollouts, AI PR reviews, rolling releases. The error rate will be much closer to 0%.


How does a phased rollout improve LLM error rates exactly?

Error rate here is the rate of shipping bugs to customers.

That wasn't what the comment you responded to was referring to. I guess it makes sense since you are kind of like an LLM with how you respond to input.

More like “Producing 80% of the correct answer” and the remaining 20% with some nudging and tweaking. Still extremely valuable.

> I feel the same way about the current crop of AI tools. I've tried a bunch of them. Some are good. Most are a bit shit. Few are useful to me as they are now. [...] If this tech is as amazing as you say it is, I'll be able to pick it up and become productive on a timescale of my choosing not yours.

I think the point the author is making is not that it's all useless, but against the very overly simplistic idea the plot of Amount of AI vs Productivity in All Situations is a hockey stick chart.

Being told to be excited about something when clearly all they're saying is "it works sometimes, other times not so much. I'll keep checking and when it's good enough for me I'll get on board" is aggravating.


> Instead of doing 1x your normal work, you can do 5x while still maintaining quality.

Yet my pay stays the same, all my coworkers get fired, and Sam Altman gets all of their paychecks. Hrm.


To be honest, I would rather spend 5x effort while doing my normal work, because salary won't grow.

> Instead of doing 1x your normal work, you can do 5x while still maintaining quality.

That's a gross overestimate. 2x I would maybe believe.

Someone has to sign off your work and unless it's hard to write but easy to read, this is where the bottleneck currently lies.


Everything is relative. If your systems aren't adapted to AI development, it will be much lower.

What if the calculator had randomness built into it?

100%. This is why I'm so reluctant to give any access to my OpenClaw. The skills hub is poisoned.


You might find your answer with `zx`: https://google.github.io/zx/



Zod is installed in nearly every project I use. It’s an essential part of my toolkit. I adore this library. It near perfect as-is and these additions make it even better. Thanks for all the hard work.


I used to use Zod until I realised it’s rather big (for many projects this isn’t an issue at all, but for some it is). Now I use Valibot which is basically the same thing but nice and small. I do slightly prefer Zod’s API and documentation, though.

Edit to add: aha, now I read further in the announcement, looks like @zod/mini may catch up with valibot -- it uses the same function-based design at least, so unused code can be stripped out.


Though exciting, looks like there's still room for shrinkage, the linked article puts a bare-minimum gzipped @zod/mini at 1.88kb, while Valibot is at 0.71kb [1].

[1] https://github.com/anatoo/zod-vs-valibot


I know this is a port but I really hope the team builds in performance debugging tools from the outset. Being able to understand _why_ a build or typecheck is taking so long is sorely missing from today's Typescript.


Yes, 100% agree. We've spent so much time chasing down what makes our build slow. Obviously that is less important now, but hopefully they've laid the foundation for when our code base grows another 10x.


Please do.


Agreed. I first seen it at Stripe (along with prefixing every ID). Whoever at Stripe (or where ever it was invented) needs a good pat on that back. It's adoption has been a huge for DX generally.


We can always work backwards, regardless of AI.


Sure, but with this new predictive model we will have better predictions to work backwards from.

OC was saying (I’m going to paraphrase) that this is the death of understanding in meteorology, but it’s not because we can always work backwards from accurate predictions.


Or we could wait 15 days and work backwards from what the weather actually turned out to be.

I guess there could be some value in analyzing what inputs have the most and least influence on the AI predictions.


Comparing the difference between correct predictions and incorrect predictions, especially with a high accuracy predictive model, could give insight into both how statistical models work and how weather works.


Cool. I'm a bit unsettled that my camera's green dot didn't turn on for it though.


Lex Fridman has a long interview [1] with Marc Raibert, CEO of Boston Dynamics, which is really excellent. It might partially or wholly answer your question.

[1]: https://www.youtube.com/watch?v=5VnbBCm_ZyQ


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: