visarga

9 ms·

181.
by visarga 4mo ago
That is how I use models too. Sometimes I have them role play the supporters of the idea I am arguing against.
182.
by visarga 4mo ago
> and that is a department we have no way to teach an LLM because if nobody is writing out those unspoken, subconscious rules then the LLM has nothing to read about them in its training data. I think on the contrary, LLM providers accum
183.
by visarga 4mo ago
I think what you are doing is good, I also have a similar workflow, but the idea here is to automate some of your manual approval work with coded tests. Since they are easy to generate, have as many as possible, think hard about what to tes
184.
by visarga 4mo ago
The centralized choke point of web search is getting relaxed now. Unlike search engines and social networks, you can download a LLM and run it. A small one, but capable of using a library of search stubs to directly fetch information from h
185.
by visarga 4mo ago
> What then? LLM learning from LLM doesn't really work, does it? It does work, it is called RLVR, reinforcement learning from verified rewards, is is based on testing code by execution. It's become a major area of improvement i
186.
by visarga 5mo ago
Wondering if anyone considered the idea of saving search stubs and using local LLM as agents with it. Could cut the search engine completely from the loop. The idea is simple: collect search stubs, short documents with routing information a
187.
by visarga 5mo ago
If you are worried about agents diverging from user intent why not log user messages in a file, and make it a point to review this file against plans and executed work? In my own harness nothing the user types gets lost. It might be the mos
188.
by visarga 5mo ago
> Especially as there has not been any new great revenue sources outside AI in recent years. Nobody can make a profit with AI. Any clever idea can be cloned with AI, competition makes it unprofitable. No moat, no arbitrage opportunity. &
189.
by visarga 5mo ago
I did the same thing - task oriented work, each task a md file. I have a harness based on it: https://github.com/horiacristescu/claude-playbook-plugin
190.
by visarga 5mo ago
AI is expanding to meet the needs of expanding AI. Why worry about jobs? AI will provide plenty of work. If anything, I worry we'll be working more, not less. All that AI will need someone to vouch for it and to scapegoat when it makes
191.
by visarga 5mo ago
I find myself observing how my lead manages meetings ... "ah, this is like when I do that with Claude", "this is where he wants to understand what happened, like when I ask Claude" ... it's funny.
192.
by visarga 5mo ago
> But let's not forget that if author cannot live of what they create, they, for the most part, won't be able to continue creating. This is an old problem. Probably only about 1 in 5 authors can rely entirely on writing income,
193.
by visarga 5mo ago
I don't think anyone's "making money" yet. We have a race to build up hardware for AI, and one to train models. There are some profits in there, but who's making money from the work AI performs? Nobody, because any
194.
by visarga 5mo ago
> "you" could be a program running on a computer If you are in a sim, then the sim execution is an expensive process, it produces heat and consumes energy. If you are in reality, a material body, then keeping alive also consume
195.
by visarga 5mo ago
> Consciousness the the fundamental reality; it is the only thing we know for sure. Who invented the words "consciousness" "reality" "fundamental" that you are now using? you are using language invented else
196.
by visarga 5mo ago
Not all brain activity is unified, I just claim it must unify on the output side - behavior and language. But this is not a brain or organism only problem. Even a cell must unify in crucial moments such as division and chemotaxis. A cell ei
197.
by visarga 5mo ago
> There must be a link back from consciousness to physics. Yes, and it is trivial to prove. One second lapse of attention and you could get bitten by a snake, or run over by a car. If the top level (consciousness) fails, the neurons die
198.
by visarga 5mo ago
What I reject most about idealism is that it is free. Life is hard and expensive, but soul is free. How come? Is our soul understood like a spherical cow?
199.
by visarga 5mo ago
There is a very good explanation here - not what brains do, but what brains MUST do. Can we walk left and right at the same time? Can we drink coffee before brewing it? No. There is a bottleneck, a serial action bottleneck on the body. So t
200.
by visarga 5mo ago
I think there is a gap, but it's not what they think it is - it is the gap between description and execution. Chalmers wants a description level explanation for the properties of execution, which is not possible. He wants to get for fr
201.
by visarga 5mo ago
Because of that I never touched Photoshop in 25 years, since PS6 I think. I rarely need to edit images, and when I do I regret the app I once had. There's no free tier or something that makes sense for my needs. I use Gimp but its UI d
202.
by visarga 5mo ago
It's no longer a choice. The AI is already causing changes in how competition, investors and users behave.
203.
by visarga 5mo ago
> How many people are just using ChatGPT or Gemini to get the image edits they want instead of reaching for Photoshop? Lately ChatGPT can generate complex diagrams with lots of text so I used them to make slides.
204.
by visarga 5mo ago
> It's just "intellectual" botox. Could be just ESL, it's hard to close the proficient to native gap.
205.
by visarga 5mo ago
Besides fill factor being hard to match, there is also scaling - you can't scale local inference 10x for a spike, but you can with cloud inference.
206.
by visarga 5mo ago
> Frontier AI companies are selling at a loss. There are huge economies to be had by batching requests and using lots of RAM for MoE (sparse models). You can't achieve that efficiency at batch size 1 on a single node.
207.
by visarga 5mo ago
Why just US and not China, France and UK? They all trained on internet text.
208.
by visarga 5mo ago
I made a harness that preserves memory for both user messages and task execution. One reason this works is related to judge agents - they can't review information that was not written down. So I track everything in my harness. The judg
209.
by visarga 5mo ago
Why not preserver the essential memories in text? Why neuromorphic?
210.
by visarga 5mo ago
> No. It should help a person develop into a free, thoughtful, well-rounded human being. That was the goal maybe in the past when only rich people could afford an education.

More ›