visarga

15 ms·

301.
by visarga 8mo ago
I'm switching from Claude Web to Claude Code. Local files give me memory I actually control, unlike Anthropic's implementation. CC doesn't carry state between sessions — you just put whatever project context it needs in a fil
302.
by visarga 8mo ago
> they call it fun, i call it death. Are you just sitting there as if dead when using AI? I find AI work exciting, always something new to discover.
303.
by visarga 8mo ago
Reimplementing function is not the same with wholesale regurgitation.
304.
by visarga 8mo ago
> Not understanding how consciousness is created doesn't make it divine. It's not divine, just expensive, and has to pay its costs. That little thing - cost - powers evolution. Cost defines what can exist and shaped us into our
305.
by visarga 8mo ago
John Good's quote is pretty myopic, it assumes machines make better machines based on being "ultraintelligent" instead of learning from environment-action-outcome loop. It's the difference between "compute is all yo
306.
by visarga 8mo ago
I use a task.md file for each task, it has a list of gates just like ordinary todo lists in markdown. The planner agent has an instruction to install a judge gate at the top and one at the bottom. The judge runs in headless mode and updates
307.
by visarga 8mo ago
You can't improve the agents but you can improve their work environment. Agents gain a few advantages from up to date docs: 1. faster bootstrap and less token usage than trashing around the code base to reconstitute what it does 2. car
308.
by visarga 8mo ago
It's our job after all to keep the agent aligned, we should not expect it to self recover when it goes astray or mind its own alignment. Even with humans we hire managers to align the activity of subordinates, keeping intent and work i
309.
by visarga 8mo ago
> This article is far off the mark. The improvement is not in the user-side. You can write docs or have the robot write docs; it will improve performance on your repo, but not “improve” the agent. No, the idea is to create these improved
310.
by visarga 8mo ago
> Nothing you write will matter if it is not quickly adopted to the training dataset. That is my take too, I was surprised to see how many people object to their works being trained on. It's how you can leave your mark, opening acc
311.
by visarga 8mo ago
Not with bare rng
312.
by visarga 8mo ago
You can automate Momo with a rng.
313.
by visarga 8mo ago
That agent has a bug that quotes from open files all the time, consuming tokens.
314.
by visarga 8mo ago
So you are saying they are like copy, LLMs will copy some training data back to you? Why do we spend so much money training and running them if they "just regurgitate text compressed in their memory based on probability"? billions
315.
by visarga 8mo ago
If you want that level of research I suggest you ask the model to draft a markdown plan with "[ ]" gates for todo items, and plan it in as many steps as needed. Then ask another LLM to review the plan, judge it. In the end use the
316.
by visarga 8mo ago
I agree time is not what we are looking for, it is maximum complexity the model can handle without failing the task, expressed in task length. Long tasks allow some slack - if you make an error you have time to see the outcomes and recover.
317.
by visarga 8mo ago
> I wouldn't call anyone producing vibe coded AI stuff a "programmer" How about "software developer"?
318.
by visarga 8mo ago
I see nothing wrong with asking a few people around.
319.
by visarga 8mo ago
Never ask an important question just once. Ask it in many ways, and on multiple models. If they don't agree at least you know you can't rely on these answers. For important questions I run 3-4 Deep Research reports (Claude, ChatGP
320.
by visarga 8mo ago
> That's not what I got. My Opus vs your Opus, which is smarter?!
321.
by visarga 8mo ago
Better to use Gemini to search YT, you can just talk to it what you want, and see the videos right there in the chat window. I used this to find well targeted music, it can also review artists and discuss about them not just play music.
322.
by visarga 8mo ago
This does not even need a LLM skill, just load the whole code up in context, so efficient.
323.
by visarga 8mo ago
I think people forget one thing - LLMs don't even need to scrape, we copy paste and put articles and documents right into their mouth, they only need to keep the mouth open. Copy pasted content is also preselected manually, might filte
324.
by visarga 8mo ago
Google ruined it, maximizing ad sales no matter the outcomes. SEO adapted to Google, Google adapted only to maximize their own profits.
325.
by visarga 8mo ago
"12+38" won't embed close to "50", as you said they capture only surface level words ("a lot of keyword overlap") not meaning, it's why for small scale I prefer a folder of files and a coding agent us
326.
by visarga 8mo ago
Great work! I like the approach of maximum freedom inside bounded blast radius and how you use code to encode policy.
327.
by visarga 8mo ago
I built a structural zoom tool, it would fit flat or tree like content into a 10K char budget. It can compress HTML, JSON, folders, zip files, logs, chat sessions, basically large files or collections of files. Moving around is done by rang
328.
by visarga 8mo ago
Make a CLI tool for it, of course
329.
by visarga 8mo ago
Already made a harness for Claude to make R/W plans, not write once like they are usually implemented. They can modify themselves as they work through the task at hand. Also relying on a collection of patterns for writing coding task p
330.
by visarga 8mo ago
Yeah I invented a similar method for information extraction attribution around 2022, I would place custom markers in a document so the extraction model can reference them together with the answer and be unique on the document to be able to

More ›