cranium

With this Fable release my goodwill for Anthropic went down.

"Generous and exciting" were my thoughts when I bought the $100, then upgraded to the $200 sub last year. Now I get big FOMO because I won't be able to pay for Fable once off the sub, and I got the global limit reset at the same time as my weekly reset (Codex coupon for limit reset feels way better btw). The all-you-can-eat buffet put the nice items into a separate menu and it won't feel like the spot to take your family anymore.

Anthropic can find ways to integrate Fable into subscriptions so you'd still be part of the same tribe, even if you only get a sip of it for a while. A complete shut off tells you: sorry, the policy is changing and this place will be about showing off your access to the best.

That's weird but what would have been a Fable-ulous (sorry) addition to the offer just made other subscriptions from OpenAI/ZAi/... more appealing, because they don't segregate people.

Wait, will it downgrade to Opus even when using with extra usage? So you pay a hefty amount up to the point where the model needs to do Real Work then it quits?

I don't understand why simonw's comment is dead, because he mentions a real counterpoint to the video: API token prices are NOT the raw costs for any provider. I'd even say that inference needs to have quite a juicy margin to cover for all the other costs. It would make no business sense to sell API tokens at a loss: nobody knows yet how to price intelligence, so why start in the red when it's the only source of revenue?

It's a different story for subscriptions. According to my rough computation (N=1), a Claude Max 20x at $200 gives you access to around $8k worth of tokens per month – but they don't cost Anthropic $8k! – and there I think they'd make a loss on every token maxxer which may or may not be compensated by subscriptions that are not used. But that's not the end of the subscription story.

Once you are "enterprise" you pay for token use and there is no way around it: Anthropic does it and so does OpenAI. The subscription is the gateway drug to token maxxing. When people are hired in an Enterprise job, they'll come with their habit of using AI for all and any task.

All to say that: yes, AI labs are bleeding money but on everything else – datacenters, training models, talent,...

For a client I had to use "Windows App" to connect remotely to their cloud machine – and finding the app was the easy part.

Nice idea but I'm missing the specialty bins that actually make Gridfinity useful: bins for storing the AA/AAA batteries vertically, for SD cards and USB keys, for a caliper, tape measure,...

With plain bins, you don't get the "this tool can only be stored there" lemma that changes how you think when you have a lot of tools. If a tool has only one place to go: 1) either it's there or it's used on a work surface, 2) it goes back there and not in a possibly-related dump (does this special double-sided tape go with all the tapes or with the leatherworking supplies?)

For other Gridfinity content:

  - Generate a specific bin for your tool: https://www.tooltrace.ai/
  - Generic bin generator: https://gridfinity.perplexinglabs.com/pr/gridfinity-rebuilt/0/0
  - Hub for links: https://gridfinity.xyz/

Electricity price is a weird beast. Everyone has to pay the price of the most expensive electricity source (generally gas plants) that was recruited to respond to the power demand. It means that during a spike the electricity price can double or triple.

What I infer from Anthropic post is that they will estimate the energy price as if they weren't using it and pay the difference if their use upped the price.

A well crafted policy that, I think, will be adopted by many OSS.

You'd need that kind of sharp rules to compete against unhinged (or drunken) AI drivers and that's unfortunate. But at the same time, letting people DoS maintainers' time at essential no cost is not an option either.

Looks slick! In the >50" category, I've recently upgraded from the Samsung Odyssey G9 49" (with res 2x1440p) to a Samsung Odyssey G9 57" (2x4K). With a tiling window manager and workspaces it's really a pleasure to use, and contrarily to some beliefs, I do more focused work that way because I don't have to switch workspace to find the information I'm looking for – less risk of distraction.

I've linked it multiple times on HN already but winapps (https://github.com/winapps-org/winapps) can be a game-changer for people relying on some Windows-only software.

It sets up a Windows machine in Docker where you can install your apps, then you'll get .desktop applications that starts the program in the VM and use RDP to only show the app window – it feels nearly native. I've even bought an Office 2024 license to improve some VBA Excel macros for a client.

It's funny how there are operations so sensitive to latency that even half a second feels too long.

However I agree with other comments that the author's baseline of 380ms is suspicious. I get 150ms (full config, 6 plugins) vs 50ms with no config and plugins.

I've found that cooking extra food with the intent of freezing it in individual portions is a game changer for when I'm alone at home - my fiancée can also pack them for lunch. Rice, curries, ragoûts are really nice to get out of the freezer, put on a plate in the microwave and eat a few minutes later.

Look at Souper Cubes (or any silicon knockoff) for the molds.

"How to be in good health? Sleep, eat well, exercise." However, knowledge ≠ application.

It will be twelve years next month. I remember signing up as a naive student attracted by all the shiny things.

I'd say HN had a big part in shaping and honing my critical thinking.

I'd add classes about:

  - telling clients that the proof-of-concept is non-conclusive so it's either bag it or try something different
  - spending innovation tokens in something else than a new frontend framework and/or backend language
  - understanding that project management methods are tools (not rites) and if your daily standup is 45min then there's a problem

Fascinating! Saw a porcupine at 0013, a fennecs at 0027 & 0257, joined by a springbok (thanks ChatGPT), sassy zebras at 0320 (you can see their social behavior!), a jackal at 0740,...

Soothing experience :)

I find coding agents to be a powerful throttle on the speed dimension in exchange of quality in the general sense. With tests to ensure correctness is not compromised (not a silver bullet), linter and pre-commit for code style, the byproduct of AI code is utter verbosity. Pleads to be concise don't work – no more than the "Don't be wrong/hallucinate". In that regard, I like Blaise Pascal quote "I have made this longer than usual because I have not had time to make it shorter."

Commit even as a WIP before cleaning up! I don't really like polluting the commit history like that but with some interactive rebase it can be as if the WIP version never existed.

(Side ask to people using Jujutsu: isn't it a use case where jujutsu shines?)

Places where I got the most out of coding agents are:

- breaking through the analysis paralysis by creating the skeleton of a feature that I then rework (UI work is a good example)

- aggressive dev tooling for productivity on early stage projects, where the CI/CD pipeline is lacking and/or tools are clumsy. (Related XKCD: https://xkcd.com/1205/)

Otherwise, I find most of my time is understanding the client requirements and making sure they don't want conflicting features – both of which are difficult to speedup with AI. Coding is actually the easy part and even if it was sped up 100x a consistent end-to-end improvement of 2x would be a big win (see Amdahl's law).

I like this lossy compression / decompression analogy for coding too: when you prompt for a feature, you are basically asking to decompress the meaning of your ask into your existing code. Any semantic gap in your prompt will be filled with plausible glue, ie. the LLM makes decisions for you. A good prompt minimizes the glue needed and reduces the potential for really crappy outcome, but it's always a possibility!

I'm working on test scenarios (in natural language) after defining my initial task to keep my mind on the same topic. Edge cases to test, implementation details to check, log traces where it makes sense,... Basically crystalizing my vision of what the code should do and look like, so I can check against the implemented version.

I find the model very impressive, but how could it be used in the wild? They mention robots (maybe to test them cheaply in completely different environments?), but I don't see the use in games except during development to generate ideas/assets.

I guess it should be "agentic" but it's used so many times it can't really be a typo...

The Walrus operator! Introduced in Python 3.8 according to the PEP 572.

Really nice to combine 1) checking if something exists and 2) act on it

I'd be happy to spend my Claude Max tokens during the night so it can "ultrathink" some Pareto improvements to my projects. So far, I've mostly seen lateral moves that rewrites code rather than rearchitecture/design the project.

Has anyone done a (somewhat) apple-to-apple comparison between opencode and claude code, as they both can use claude pro/max subscription?

I'm curious about how they feel to use and their "performance".

Instead of "pretty trivial", I'd say it's "well-defined and generally understood".

The implicit decisions it had to make were also inconsequential, eg. selection of ASCII chars, color or not, bounds of the domain,...

However, it shows that agents are powerful translators / extractors of general knowledge!

That's a really good idea! That would handle micropayments that nobody would even bother with (to pay, to process, to receive, ...).

Could even have users select the payment %age or have it set by the contract tier between the app creator and the user (10% for simple user, 20% for pro access with other features, 40% enterprise,...).

Looks like the GraphQL beginnings, it feels fresh now and the AI products kind of nudge everything that way but fundamentally the main point is being able to package an API into a standard format, with documented endpoints, support for authentication, and different transports.

Nothing that couldn't be solved by a well designed REST API with Swagger documentation (and authentication, and websockets), but by being the same interface for all APIs it reduces the burden on the LLM.

That's a problem for products that have been vibe coded to the bone. They may look slick, but behind the scene they can be ugly and fragile.

The accompanying video is really nice: https://youtu.be/vfgySTaM1TI?t=269