There was ages ago a project called monkeysphere that let you sign the host ssh key with your gpg key and verify it automatically. The downside was that it was very slow.
Yes, and this is obvious if /users/ exists and returns a 400 if the ID is required. That way you can tell the difference between /users/ being there and expecting and ID, and it not being there.
I think the point is that you can have arbitrary website read the browser’s memory so example.com can read the password for example.org and example.net.
There is an LLM API. You send it a system prompt and the conversation history. If the last message is a user message the agent will send back a response. It can also send back a “thinking” message before it sends a response and it can also send back a structured message with one or more function calls for functions you defined in your API request (things like “ls(): list files”).
The harness is the part that makes the API calls, interacts with the user, makes the function calls, and keeps track of the conversation memory.
You can also use the LLM to summarize the conversation into a single shorter message so you get compaction. And instead of statically defining which functions are available to the LLM you can create an MCP server which allows the LLM to auto-discover functions it can call and what they do.
That’s the whole magic of something like Claude Code. The rest is details.
I would rather people who find this kind of stuff pad their resumes and get coolness points on HN than sell this exploit on the black market. But your priorities may be different and you might prefer they do the latter.
The problem is that the difference between a low tech and a high tech diesel tractor is mostly emissions and some loss of efficiency. The difference between a low tech and a high tech electric car is a 25 mile range and a 250 mile range, a top speed of 35 mph and 100 mph, carrying capacity and so on.
I recently did a lawn tractor conversion from gas to electric and what I got was in my opinion significantly better and more reliable than a commercial option at 20% of the price but it is limited to 4mph. Scaling it to 5 would require a lot of custom fabrication and a much more expensive drive motor. Once this tech is significantly better and cheaper to the point of being a commodity it will be a different story. For now it just isn’t.
What a way to ruin goodwill with the very community they are trying to court. I am on a Pro subscription to use with Claude Code, but it sounds like the days of using it are numbered. I guess I will be trying the latest offering from OpenAI and Google tomorrow and if they are satisfactory I might just switch. Moreover, I have been recommending Anthropic's API solutions up to now to friends and clients. Based on this dumb move I will be now starting with this anecdote and then giving a very hedged recommendation.
Realistically the future of all this is that open models become good enough that LLM as a service becomes a commodity with a race to the bottom in terms of cost. Given where we are today I can easily see open weight models in 2-3 years making Anthropic and OpenAI irrelevant for everyday development work (I justify this like so: if my coding agent is 10x smarter than I am, how would I understand if it did all the right things? I want someone of roughly my intelligence for coding. I can see use cases for like independent pharma work or some such where supergenius level intelligence is justified, but for coding ability for mere mortals to reason about the code is probably more important).
The valuation is obviously based on the premise of their capturing the white collar economy. OpenAI's charter says so openly. And Chinese robots will come for blue workers next.
The economy, not the workers :) It feels like pretty soon white collar workers will be in a “You have nothing to lose but your chains” situation. Except we are not as fit as the proletariat of the past.
In my experience, Codex is better than Claude Code in every way and GPT-5.4 is on par or better than Opus 4.6 at every coding task I ask of it.
You're really not going to miss CC. And OpenAI actually had some foresight to invest massively in compute so they don't run into usage and rate limits like Anthropic does constantly. I couldn't even use CC for more than a couple complex tasks before I was out of extra usage or session usage. It was a maddening productivity killer and I just switched to Codex full time.
I am on Google's $20/month plan, and I usually get about three half-hour coding sessions a week with AntiGravity using the Claude models. The limit using Gemini Pro models is much higher. I am retired so Google's $20 plan is sufficient for me, but I understand that people who are still working would need higher limits.
I am also on a $10/month plan with Nous Research for supplying open models for their open source Hermes Agent. I run Hermes inside a container, on a dedicated VPS as a coding agent for complex tasks and so far I find the $10/month plan is enough for about five to ten major tasks a month. I think it is also a good deal.
If I could get the equivalent of GPT-4 running locally, that would cover like 95% of what I need an LLM for. Tweak this dockerfile, gimme a bash script. I guess the context probably isn’t sufficient for the agent stuff, but I’m sure more context-efficient harnesses will be coming down the line
I have an old Mac Mini with 32G of integrated RAM, and the following works for me for small local code changes:
ollama launch claude --model qwen3.6:35b-a3b-nvfp4
In addition to not having an integrated web search tool, one drawback is that it runs more slowly than using cloud servers. I find myself asking for a code or documentation change, and then spending two minutes on my deck getting fresh air waiting for a slower response. When using a fast cloud service I can be a coding slave, glued to my computer. Still, I like running local when I can!
I don't really buy that. There are a lot of situations (e.g. being directed to park in a space at a fairgrounds, ski area, or whatever) that you can't reasonably expect AFAIK to be programmed into a car's computer. Even if a car can legitimately handle roads under most circumstances, they're not going to be able to handle everything.
"Because the Origin does not have manual controls, the NHTSA must issue an exception to the Federal Motor Vehicle Safety Standards to permit operation on public roads"
There is a reason that pilots get basically told the ins and outs of a specific plane. Imagine the outrage if people need to do month long training for a specific car just to be able to drive it (and not just a general "here is how cars roughly work and the laws of the road").
Airline pilots aren't supposed to take a nap, and there are occasionally articles about the various things that have gone wrong because the pilots weren't paying attention.
How do you reverse such a car into your own driveway that's positioned in a funny way at an angle and an incline? What if you're parking off road for any reason? Like, you have to be able to manoeuvre your own vehicle sometimes.
The simpler explanation is that an industry insider who can publish a piece saying “helium shortage will mean the end of chip making as we know it” can get a lot more views and clicks than one who published “chip making will get mildly more expensive because one of the key ingredients is going to need to be sourced from farther away or from more expensive suppliers”. There is always an angle, whether it is clout, pumping the market, selling you something, etc. and when you are not an industry insider there is little you can do to understand where else you can buy the particular ingredient from so it sounds plausible.
reply