justinhj

I like the idea. It doesn't work well in the happy case that your baseline is 300ms and you get a 20x speed up, but I guess you can adjust at that time.

I also incorporated some of the authors previous ideas into my benchmarks: see https://matklad.github.io/2025/12/09/do-not-optimize-away.ht...

We can talk to computers now, and they can both understand is and take actions. I expected things to move much faster to the point where 99% of what people use tech for is handled by speaking natural language to pretty much any device and an agent does what you want. Maybe it's coming but it's very slow, or maybe I am way off.

That seems a bit unfair. The choice isn't between opinionated vs unopinionated, it's "which layer do the options live at?". The harness has to be somewhat opinionated about the implementation of the agent loop, otherwise what is the harness exactly. I guess the question is if I don't want the new "code mode" changes in the loop can I trivially ignore it, but it seems it has to be implemented in the core.

Prompts I am not that sold on but it seems silly to not implement it in clients.

A good use case for resources is small amounts of commonly needed state that can be fetched and proactively updated by the mcp server, saving latency when the model requests it.

So it's obvious we will not have a conventional war

That's what I call a Turing test

Russia is a nuclear power

The Geico ad gave me a laugh.

I think Google realized this first.

Yes, it is.

In the US "Could care less" is the equivalent to the British expression "Couldn't care less". The intention and meaning is the same. So the original comment did mean to say what they said.

My point was that what may look like a mistake or a contradiction is just cultural development of natural language.

Another example is the British expression "It would be cheap at half the price"

Countries have different idioms.

We should start to discuss things like this vocally, in meatspace.

vcpkg is pretty good. cmake is complex but very flexible. just need an llm to help and it's pretty much solved (if your dep is in autopkg)

Same fear. At least it wasn't Google

They have simply been living outside your bubble

As a former game coder who worked before there were game engines the attitude depresses me. Game programming was an inventive and creative art and the models that are spewing out barely playable imitations of games are using all that code we put out there, just as they are using all the art. Same is true of game designs.

I think we should be making games with AI however. I just don't think art is special; we have all been robbed and now we should all use the new tools that are the result of that robbery.

The author is right, the website is nice, but I am having trouble finding my star sign.

They deferred navigation to ATC instead of using their own capabilities such as VOR.

At the time Starlink was announced as was rocket reusability. Both were widely mocked and considered infeasible.

Since then Starlink has succeeded, reusable rockets are a reality, ISS crewed spaceflight restored to US.

I expect all that had an impact on the valuation. Personally I wouldn't invest in Elon because his stock is a wild ride but space x has achieved a lot.

If the AI tools can easily find exploits, cannot those same tools be used to harden security? In fact the companies have an advantage over script kiddies: access to more expensive models and compute time as well as professional security engineers

you live in a bubble i am afraid

if that's all he did sure, he makes an interesting blog nearly every day jealous or what?

I wonder if fiction writing at that level is like coding. You can learn to type and navigate really fast, but unless you're working on blub software, if you're doing something interesting and novel, you won't actually get to the goal faster. The bottleneck is the ideas, the structure of the story and the development of the characters, amongst many other concerns.

I find Simon's work informative and entertaining; the last thing he can be accused of is low effort. The Pelicans are just a bit of fun icing on top.

lmao

it's worse really. at least there are subreddits relatively free of trump bad elon bad non stop nonsense

are you sure it's non trivial? they posted a 2d image of what it looks like. a fairly simple model of the users wrist and mouse position doesn't seem crazy hard but the devil is in the details

There's a good book "The Pacific Alone" about a guy that did this in a kayak

Unlike Uber and Airbnb that did break local laws and got away with it because people wanted their service (and also deep pockets for handling litigation and encouraging politicians to see your way), training an ai is generally not theft.

If I read a physics textbook and now I know some physics do I perform a theft when I use it practice or teach someone else?