wood_spirit

My experience - in the Nordics, but this is transferable - is they are also much nicer to own and drive. I don’t know anyone who went electric and then went back. The generation who went electric a few years ago is now on their second or third electric car. The makes and models in favour is changing (cue tangent on Tesla brand death in Europe) but the power source choice is not. So as more people get one, I expect their market share to just climb and climb.

As soon as the competition is bankrupted they no longer need to release for free? It’s like how big players enter markets by launching at a loss to destroy competitors?

Rather than DSLs I’ve found careful force tools results force the same kind of discipline in a more straightforward way to implement. So it’s normal to ask the llm to answer only yes or no and they are pretty good at following that instruction but it doesn’t scale so well. Whereas if the shape of the force tool call gives them more richness without giving them freedom to go off piste it scales to more nuanced results whilst also being trivial to parse.

For a lot of people it’s not the job that is rewarding it the role having a job gives them in life and home life. Financially contributing to the household through earning it through work is a meaningful and rewarding thing that can define the near total of how good you feel about yourself thing even if you don’t like the job?

Another ancient board game to be decoded is https://en.wikipedia.org/wiki/Royal_Game_of_Ur. This is from 2600BC so is older to Christ than Christ is to us.

The videos on the game (and all his other videos) with Irving Finkel, a curator at the British museum, are spellbinding. He has the looks, manners and enthusiasm of an eccentric museum curator from central casting!

This looks staggeringly brilliant.

I was searching for something like this recently but ended up rolling my own using a block-based OT to local server and diff sync to remote ones.

I’m reading the system guide and nodding along. It’s super validating to see the similarities and contrasts.

Countries can be repugnant in different ways, but the trust in their word is completely orthogonal.

Can a country trust a trade deal with the US? Whereas when has China ever not honoured a trade deal?

This is not about whether being a customer of the us or china is morally worse. It’s about trust. Does any country trust America at its word any more?

I am a careful follower of politics and don’t think I’m ignorant. And on every point you made I could rebut with an example of America doing likewise.

On your last point abour “causing instability on the internet”, for example, I could point out we are in a discussion about America wanting to restrict LLMs into haves and have nots and if that isn’t destabilising the internet by preventing non-American companies from defending themselves or paying American companies to defend themselves against them protection racket wise then what is?

As the US jerks from being BDFL to bullying zero summer, it’s not surprising that the abstract fears of china seem manageable compared to the new pressing fears of the US.

Why as a European be more afraid of a country that whilst you wouldn’t want to live in and which has supported Russia is quiet and selling you cheap stuff vs a country who used to be a friend who is now wanting you to grovel for a bad deal and is also supporting Russia?

Expect the US to sanction non-US-controlled models and put sanctions on individuals, companies and countries that use them? They already do this with other things like oil.

Beautifully written but when the lack of a better compiler gets attributed to rational actions my brain glitched. That’s not fitting my mental model of how big corps operate at all!

Occam’s razor IBM didn’t invest in Fortran I because the internal political environment at the corporation didn’t have the incentives aligned to do so. This is completely orthogonal to whether they could have used a better compiler or not.

Middle managers were about making work legible https://www.seangoedecke.com/seeing-like-a-software-company/

Now that wasn’t the most efficient and it has a whole bunch of perverse incentives too, but it’s what happened as companies got big enough to hire people to do the work that could be delegated because those delegatees got status from having people to themselves delegate to.

This completely chimes with my own thoughts on this. In my mind a lot of AI adoption has been about AI Efficiency” - ICs using AI to do what they’ve always done with less effort but no actual change to the org chart. Whereas I think we need to think about “AI Effectiveness” - using AI more smartly to better communicate and coordinate what we are doing, and transforming the org structure and ways of working towards “AI Native” - the North Star that a fresh competitor startup would execute when starting out with AI as a given.

Zigzag encodings are a common compression scheme used in the Parquet format. It is fun to speculate that these kind of tricks could be applied there in something so commonly under the hood of a lot of data processing and analytics

The articles and discussion around this and the Amazon story all seem to me to be an earnest tech press and community searching for a genuine reason for the administration blocking Anthropic’s models.

However, thinking back to the spat with the DoD and more generally how the administration is much more supportive - and supported by - OpenAI and XAI and it’s easy to imagine this is just another escalation in the fight between a “liberal leaning” company and its competitors and the administration.

There might have been something said by someone at Amazon or something but I’d guess Occam’s razor the administration just leapt at the chance after their supplier sanctions fell flat?

In my own multi agent framework I use cheap models to check the responses of the expensive models, as well as using multiple expensive models adversarially in debate. The cheap models are great at spotting eg the model getting stuck in the alternate between two broken ideas or not following code conventions or missing a step in the skill and so on. I’m currently working on making them detect user corrections and police that going forward to intervene when the expensive models forget the thing you just corrected them about etc.

Yes context management is key.

I do my own framework and spend a lot of time trying to debug this and it’s not so much the context size in hard numbers but rather the probability that there is debris or wrong directions in the window that are drowning out the things the user thinks are important.

This manifests in the llm that keeps going back to doing the thing that failed when they tried it just before the last approach etc. The frequency of things in the context window give weight even if they are the wrong things.

I have a lot of tricks like not giving the llm lots of tools but rather giving it a tool it can use to search for tools etc.

But the bigger solution is in process where you use something like superpowers to force the llm through stages and you control the context that carries forward.

For me turbo pascal - with inline assembly - was the pinnacle. I got into c and later c++ because I had to, but always found the symbols slightly harder on the eyes and surprisingly not faster to type. And I was always frustrated by the bloat of the executables and the much slower compilation times. And the runtime speed - I was doing a lot of assembly, it was something I became interested in even on projects that didn’t need it - was actually much faster in TP. It was, in my eyes, the perfect blend of easy on the eyes syntax, blazingly fast compilation and runtime and small easy to share executables.

Then of course Delphi came along and made all that true for windows apps too!

So somehow I chime with how your comment starts but have such different memories of how it ends :)

The real question as xAI has just made Musk a trillionaire is how this, the most recent blow in a fight between the administration and Anthropic, impacts Anthropic’s IPO.

I’ve used Fable a lot. It’s a marginal improvement on Opus. It’s really not scary smart. I don’t want to go back to Opus because Fable was that little bit smarter, but it’s not like anything really changes at work and when we’re bumped back. So I’m really not buying any national security angle other than in the sense the administration can weaponise that to crown the AI race winners.

Related post from same site earlier this week https://news.ycombinator.com/item?id=48385457

Of course we can hope for more, but would you agree it is a good start though?

Same! And there is no obvious close button etc on the overtop blocking modal dialog either. It’s a dark pattern.

As soon as I saw the beautiful rendering I wondered if it was using spherical geometry so I asked an AI. It dug and found discussion on d3 from the author about the algorithm :)

https://github.com/d3/d3/issues/1820

Beautiful :)

The idea that springs to my mind is to do Delaunay and Voronoi using spherical geometry. I think the article uses flat Euclidean geometry but if we tweak the fifth axiom we could do spherical or hyperbolic?

If you need (or want the convenience of) a uuid and the time of creation is not secret then use ulids eg uuid v7.

I have my own llm wrapping harness, which does this and has a few more tricks. For example, it doesn’t have a lot of mcp but it does have search_mcp and load_mcp tools (and search_skills) so the llm can find what it needs when it needs it without bloating the normal baseline context. The LLMs have proved really good at using them. There is also a waypoint tool they can use to record their thinking in the context without it being the final output. Am thinking about a search_expert to find colleagues it can bring into conversations too. And a lot of other stuff.

Pro tip they worked well for me with response truncation: in the truncated output, say that the full text is available in /tmp/whereever.txt - that way, the llm will be able to query and read more using built in tools without reissuing the big tool call.

This is like so counter to how it feels on the ground. I genuinely don’t know anyone in Sweden who has cash or wants cash in their pocket anymore.

Reallocating human time is also going to cause problems.

But it’s a great short term business opportunity for AI vendors and it was Anthropic who went all in on being knowledge worker outsourcing in a big way first whilst OpenAI thought they’d replace Google in search.

I think Anthropic had the better business strategy.

The US is a country with lower hunting participation than many other western countries and yet an order of magnitude more gun owning households.