Hacker Newsnew | past | comments | ask | show | jobs | submit | aix1's commentslogin

The UI is full of glitches: the legend that's placed right on top of Australia, the title that doesn't fit in the box, the crosshair that doesn't accurate track the cursor, the pixellated fonts along the perimeter, the unreadable colour combinations in the overlay, the rendering glitches along the axes when you flip from tab to tab and so on and so forth.

It's like someone took a beatiful, intricate piece of vintage jewellry and made a slapdash imitation out of cheap plastic.


Yep. People are creating garbage with AI that looks passable at first glance, or maybe acceptable if you have no taste. This is the kind of software we can expect to receive in the next few years.

> And this article is basically = you will get a road, built quickly

That's not how I am reading it. You will get a road built exactly to your spec, quickly. So no penguin crossings unless you ask for them.

I am also not entirely sure how the pothole argument translates.


The road will be built to some specs, including features nobody asked for. If the corpus was trained for roads built in Arctic, you will get penguin crossings.

Did not trigger for me (Fable answered the question), so I guess the filters are either non-deterministic or are still being tweaked.

Interesting, I assumed all model-routing was done utilizing an LLM. (I.e. non-deterministic.)

It’s possible that there’s a set of words or phrases that route deterministically to save money on obvious stuff.

I kind of wonder, though, which model they’re using to do the routing. It seems like a huge added cost to do these kinds of checks on every request


Wasn't it leaked in the Claude Code source that it was all regex?

> It's the keys to a substantial chunk of the kingdom for $1B a year. Literally they are getting, for a very small price, the right to distill their own models from Gemini.

Here is a different interpretation: Apple bought the rights to distill and use a smaller version of one unspecified model in the Gemini family (there are many such models).

The distillation will be carried out at Google's data centres so that the original weights never leave Google premises.

For this to be keys to be kingdom it would need to cover all current and future models and would need to be very permissive with regards to distillation parameters and allowed uses of the distilled model.

I expect the reality to be somewhere between these two extremes.


Funny, yes, but IMO a terrible idea. :)

It would help the writer once, but impose a cost on all future readers for the lifetime of the code.

It's a bit like reading English with bits of German, French and Russian. All of sudden you have to know that Buch, livre and книга all mean the same thing.

Not to mention that there are often subtle differences in meaning between words that on the face of it seem equivalent (in both human and computer languages).

It could be a nice feature for an IDE though, to help someone learn a language.



Probably Recursive Self-Improvement



That is not quite the right word. For Python, the headcount was moved from the Bay Area (the most expensive place in the world to hire software engineers) to Munich (the most expensive place in Germany to hire SWEs.), for cost saving reasons.


If you're the author of https://github.com/hanwen/go-fuse/ -- thank you :D


The problem is most SWEs in Germany are not as good as most SWEs in the Bay Area. :(


Most of the engineers making most of the tools being praised in this thread are in Germany, so I don't think that generalization quite holds.

Even if the best SWEs are better in the Bay area, there's also a lot more competition for them, so Google in Germany might be able to get top 1% there (and in neighboring countries) but Google in the Bay Area is probably having a tough time getting even top 10%.


That's a good point, and why I'm happy to see remote offices pop up in many locations. The problem is the top .1% which can live anywhere, is often a poor representation of the depth of talent density.


You're not missing much -- Claude is a better model for coding. That's what basically everyone at Google DeepMind uses and what I expect most other Googlers would choose to use IF they had access:

https://www.businessinsider.com/google-deepmind-ai-tool-divi...


I am Not a googler, just a very good google user but hope those Googlers using them third party service providers' (other than Google) LLMs read this (manually critically thinking, not skimming via a LLM layer, losing structural human nuance) :

https://www.oreilly.com/radar/dont-automate-your-moat-matchi...


I've been coding professionally in Python for about twenty years (alongside, at different times, a dozen or so other languages).

I find that Claude can write great modern Python more or less out of the box, with minimal style guidance from me. I do have to nudge it from time to time to not do silly things, but overall it's really rather good.


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

Search: