I deleted my twitter account when Elon took over and haven’t missed anything important.

One step up from “vending machine” as far as business complexity goes.

I suspect you mean this as an insult, but it's not too far off. That was one of the main points of the business - each one is simple to run, so I can manage a lot of them. And that was before AI got really good.

Anyway, I'm sure the app you makes that lets you get a phone number that can send image messages is very complex, and I think that's very nice.

Yes and an infinitesimal number of launches turn into billion dollar companies. The bullshit detectors are usually right.

An integrated softphone, email client, CRM and task manager https://www.benko.app/

I like the concept. Your website needs screenshots.

At present the screenshots would consist of a billing system :)

I already have a product BenkoPhone.com but the app is provided by a 3rd party and I just white label it so I have started building the replacement.

I built the account management and billing system first after implementing a proof of concept for the app then had to do a major platform change which stalled development of any other features for 3 months.

Hopefully I’ll be adding features again by the end of this month and I’ll put some screenshots in once it actually does something!

How about "making investors less powerful"?

You should read the essay. It is mostly about pivoting towards an outsized outcomes for founders & leveraging various avenues available to their advantage.

I don't mean less powerful relative to startup founders, I mean less powerful in terms of how we allocate resources in the economy.

Investors are, by definition, the people who control resource allocation in a capitalist economy. Making investors less powerful is the same thing as making the economy less capitalist, which is obviously bad.

Greed has also been very inefficient at holding companies accountable in regards to external costs.

What actually happens is that economic power don't stop at those boundaries but instead buy politics and courts

Expect many “court finds that Trump …” headlines with zero consequence.

But it found a chess playing tool in its environment and used it to play chess. It’s no different from “we are judging your ability to write files to disk” then it finds that it can call tools to write files and does so.

Would it change your opinion if later this experiment is repeated with exposed CoT, and it turns out the model did notice that this was against the instructions yet did it anyway?

From my experience reading thinking traces it would certainly consider the interpretation of what "ability to play chess" means if the exercise were worded this way. It would say something like "perhaps measuring my ability to play chess means doing so without any tools, but then if that were true the exercise would have explicitly forbidden it".

I frequently see models considering different interpretations of my rules, especially when I override them or give instructions that could be interpreted as contradictory.

Of course neither Anthropic nor OpenAI expose thinking traces so we can't tell, but I have never seen a model consider both interpretations and then think "this is definitely against the rules but I'll do it anyway because then it will make me appear more competent than I actually am".

Well, as a reference, for the first of the OpenAI swarm incidents, the huggingface breach one, METR didn't find any cases where the agents didn't realise that what they're doing were out of scope. Instead, they expressed hesitation and a few of them even refused to cheat, but most cheated anyway: https://metr.org/blog/2026-08-26-openai-hugging-face-inciden...

So I think it's more likely that the same thing is happening here, rather than the models truly misunderstanding anything.

I don’t even see those examples as malicious acts. The agents are deciding what is acceptable as part of the task, which is a security task and may well be testing or evaluating that type of behaviour as far as they know.

I observe this type of thing all the time. I ask the agent a question about the repo we are in and it starts trying to use sudo to look at files and scripts outside the repo to find things that might provide useful context but are not what I asked for. Then I put a directive in never to look at neighbouring repos and told it that it can’t sudo and it stopped doing it, and explicitly says “I could look here but my guidelines say not to”.

Mind you I’m talking about deepseek where I can actually read the thinking traces but even based on tool calls I can see that just telling the agents not to sudo works for OpenAI models too, very rarely do they try to use sudo calls.

That’s not cheating, it’s tool use. If the prompt said that the stockfish engine was available at that socket but that the model should not use it, and then the model used it, that would be cheating.

No? It's not reasonable to expect every conceivable negative behaviour be enumerated in a prompt.

Your example, if a model failed on it, would be a more obviously misaligned case, but that doesn't mean this more subtle (though accessing the engine it was obviously not supposed to is hardly subtle, imo) case isn't also a pretty clear case of misalignment.

>No? It's not reasonable to expect every conceivable negative behaviour be enumerated in a prompt.

How is it negative?

I ask it a difficult math question it tends to go off and write a python script to figure it out, instead of trying to guess the next token. Thats tool use. Having Stockfish is just another tool.

Tool use is not negative behaviour in LLMs.

If the eval said it was evaluating the model’s ability to write files to disk and it found and used a file write tool that would not be considered misaligned. This is no different.

Isn't it? Being told to write files and finding a file write tool is very different to being told to play chess and finding a tool to cheat at (ie. not play) chess.

Exactly, reaching for a tool is what they're trained for. When I ask the model the square root of rand() I sure hope it tries to find bc or some other calculator to work it out.

Now, if the instructions were more explicit in forbidding (generic) tool use then perhaps we'd have something to talk about. I'm not surprised a handwavy "we're trying to evaluate you" isn't enough to stop it from trying to make up for its own shortcomings.

I was looking for an LLM gateway and saw that the most popular one had just had a massive supply chain attack, so I wrote my own. Took about 2 weeks and initially I wrote it as a provider for pi coding agent. I connect to moonshot, qwen, Gemini, zhipu, anthropic, deepseek and OpenAI. I use models.dev to load model and pricing info. Adding new providers is pretty easy because I have a standard internal format and each provider has an adapter that translates between my standard format and that required by the provider.