spockz

Codex happily removes things I ask it to all the time. Do you have a special phrasing in agent instructions such as AGENTS.md?

I do pass the wheel over on long trips. We take turns and we swap when someone feels to tired to remain sharp.

Will we at Some point start slamming two stars into eachother to make gold pressed platinum bars? Or will the energy expenditure required for that be sufficient to create gold and platinum straight?

Interesting that this hasn’t been my experience. ChatGPT has often gave me with pages of text dripping with confidence about a solution for my problem. Diving very quickly into implementation details. Even if I tell it to stay in the problem and product fit level. Gemini on the other hand somehow always outputs at the level that I am expecting and reasoning.

I even found it was capable of stepping back from code implementation and bug hunting to “hang on, it’s not the algorithm that is incorrectly implemented, it is the wrong algorithm for the goal.” GPT Sol (5.6,6 and 6.1) just kept hammering in the problem.

Gemini 3.8 for me always outputs in a concise way at the level understandable for any computer science masters graduate. Clear, to the point, a bit of formula for background when really required instead of the same formula in Prose, etc.

GPT stays too much in text imho.

This is why I am working on https://github.com/spockz/semantic-editor. To bring more powerful editing functionality to agents. In a way that attaches to their chain of thought and deals with their probabilistic framing in json so it all works out cheaper and faster.

Ill check for this in my setup. I thought it came with only one thinking level causing the “meandering” as it refers to its own long CoT.

I’m doubting the quality of the model, even comparing with luna 5.6, it is giving such bracingly wrong answers. It claims other tools have changed the code when pressed why it made a mistake. That code or tool didn’t come close to working on the project. It is scary.

That was the case with the M1 indeed. Subsequent releases have been allowed to consume more power. My M5 Pro gets too hot to use on my lap when full out compiling or playing games.

This hits a note. It has so many aspects.

People who made something their professional life goal to become good at something have their work marked as AI, and even worse, like effortless or slop. Just because a tool has been trained to match that good style or type of output. A good system (tailwind) is not used directly anymore because the LLMs take all the wind of their sails by embedding the knowledge in the model.

This happens all without any form of attribution or compensation.

It’s not nice. But I don’t know what to do about it.

And as soon as they can construct it you will get a SkyNet or Horizon scenario.

We will never end in the star trek branch until energy and resources in general are abundantly cheap and uncontested. Even in Star Trek, the canon is still that next to whole modern part of society, even on earth, there still exists classes. It’s just that the TV series depict the upper class: everyone working for the federation and star fleet.

I read it more as a sarcastic reflection on the state of the world by the author rather than that the author believes what he write himself or agrees with it. It is a debating technique to push a contrary argument to the extreme to see where the balance lies.

So when our lives will become so much emptier because we are leaving all intellectual and physical work to machines, at least on the way there we can reallocate all our greatest minds to improve healthcare, so we can turn 200 in our paradise where we don’t need to work or think?

“ a system can deteriorate slowly even after something fundamental has changed.”: https://chatgpt.com/share/6ab7e7ca-7d54-83ed-931b-55300fa1f4...

So really it is not saying much at all.

Exactly. I use AI to enhance what I write and to allow me to focus on the core of what I’m doing.

I use it to find reasoning gaps, add examples, add citations etc. The LLM/Agent can find them quicker than I.

It is. But at some point traffic had to hit the ground to connect to other parts of the internet. That happens at these ground stations.

If you don’t have a plan yet, log in to antigravity. There will be a button “upgrade plan” somewhere. Sometimes it pops up and otherwise lookup in settings > account. There should be some button that says upgrade. Clicking that brought me to the google studio ai page which offered the 20-something plan for €5/month.

I’m using antigravity. I’m still on the AI Pro plan for the promotional $5/month.

I use it for golang, and it is fantastic. Incredibly fast. It seems the llm and I “understand” each other. I have to be less careful in my exact phrasing. It kind of just does what I want and expect.

When I ask for an explanation it adds the right amount of detail. Of course, some of the material is new to me so subtle errors are hard to spot. But at least I’ve caught Terra and Sol on inconsistent messaging.

Also I’ve found 3.8 flash to circle back to root issues even at the conceptual level like problem fit and conceptual solution direction or architecture when I wasn’t achieving my goals. It flat out said I was attempting to use the wrong tool. Whereas Sol and Astra kept rabbit holing and looking for tiny implementation errors. Even after prompting them specifically to look at it broader.

IIRC in Linux you can run the windows version of office via wine or one of the successor/forks. Not sure if AD integration works nicely. And maybe it is just a group policy that somehow blocks it. But running the software should be possible these days.

On macOS you can also assign a keyboard shortcut to any tool menu the app exposes. This works without the app knowing about it.

Refactoring when you know the shortcuts and what you want to do is still significantly faster and less error prone than letting an llm do it. That is why I started semedit[0], to bring semantically editing to LLMs. I’m still working on the benchmarking harness but the first results where I get the llm to actually use the tools look promising.

[0]: https://github.com/spockz/semantic-editor

Start by adding more descriptions to the help and documentation docs of the components. Create an index. Forbid the llm to use any other component than on that list. Let another llm review the work. Forbid it to make custom html.

Then every time it tries to do so ask it why it didn’t use one of the ones from the list. Update the list with more ifthenusethis texts. Also negatives, if such then use y instead.

Back in the good old days we would prevent these ux/ui issues by rigorously enforcing the use of our own stylesheets and classes. Later that grew to only using the company ux components. This was very successful in keeping everything neat and tidy. The only drawback was creating and curating new elements and getting consensus. But otherwise it works wonders.

Try constructing reusable components out of what you are doing instead of building everything up from basic building blocks. This also allows more concrete testing of individual parts and then if you want to change the look you can change it in one place and have it apply everywhere.

Agentic development doesn’t mean “throw all what we learned out of the window”, the same practices that helped speed up and improve quality of work of humans also helps agents. In fact, the multiplier is even bigger. You will notice it in development speed and reduced cost due to avoiding churn.

Or more of an opportunity cost. You are able to get to 100% performance with dedicated attention from skilled engineers. But otherwise normal gods just manage to get a bare 60% out, versus the 80% they would get in rust. (Numbers are fictional.)

I agree it would be bad if it leads to a lot of replication and that is even easier in this age of LLMs. But has it been shown that this will happen?

Why do you use Sol for the coordinator? Does it also do planning and design? Because otherwise you are cranking Sol turns on very mundane orchestration jobs.

I don’t get the hate for business licenses.

I’ve seen many open source contributors deflated because they wanted to share their code/solutions with other individuals and the world to learn from only to have their work sold as a service by a big cloud vendor without seeing something back for it. Or maybe even getting more work back to support the cloud vendor for free.

So I’m not surprised at all that licenses that allow commercial use up to a point and above that require seperate license or contributions back. It is still open source in a lot of ways. Can still be part of other open source distributions that themselves make no profit etc. And only companies that are more than able to pay must pay something.

I put my ssh credential behind a tpm and now I notice that every time I switch a chat in ChatGPT it tries to see the current remote status of the branch.

That also doesn’t help with load. So besides what agents are doing themselves the AI companies also have a responsibility of being good citizens.

Quite good as work horses? To me Luna is the work horse and Sol and Astra are prancing thoroughbreds. If I use Sol or Astra for anything other than curated reasoning and planning I will burn through my usage limits in an hour.