swframe2
swframe2 on Hacker News
I want this for the visualization of very difficult to solve application evolution.
This is for problems which have no known solution and are too complicated for an agent to figure out on its own.
My current problem: can alphafold and related tools figure out the function of a gene that so far is unknown.
Think of the monte carlo tree search in the alphago explanation videos.
What if there was visualization that showed how the policy and value models worked so you can spot their flaws.
I want the agent to build an attempt at a solution,
then build a visualization of it,
then run the solution and show me what is it up to.
I want to it pause and explain its state so I can see exactly where and why the solution fails.
We need increasely difficult bio-agi contests.
<Naive>
Run the code you want to protect in a cloud function.
Cache the user data on the server; modify it on the server,
send the diffs to the browser.
</Naive>
<MoreNaive>
Any product that agent can generate from a prompt or reverse engineer will be cloned.
</MoreNaive>
<MostNaive>
Solve problems that make your life better even if cloned.
</MostNaive>
James Bond villain system activated:
1. figure out how to make money by breaking government in a way no one can agree to fix
2. pay politicians to make it legal
3. run political ads claiming the other side is the villain
4. buy that island you've had your eye on
Agreed. Novel would be very impressive. AI tech is struggling with that.
It is important not to wait for today's AI tech to get to AGI. It is better to exploit the useful things it can do today while waiting for it to get better tomorrow.
Cloning existing software might be useful today.
It allows the model to
1) learn from its mistakes
2) mix known solutions (embed product A inside product B)
3) mine value (turn product A into a set of reusable libraries)
etc.
Either way, I suggested you try to build and launch 1 product per month.
For example, someone launched a SAAS for excel formulas that now makes +$2M.
It's my understanding that when you purchase content you get a license that is limited to terms the seller defines (i.e. only if the content is delivered from seller's approved data sources, only if the content is viewed on a seller approved device, only if the content has a seller approved drm, etc).
I can't find any references to the company online anymore. The product was a c++ interpreter (created from the founder's compiler class project) and IDE. This was before java/python/javascript/etc. The interpreter would catch errors and you could mix interpreted code and binary code. Pure software displaced it.
In the late 80s, a friend turned his comsci class project into a product.
The company was making about $25m. It died due to a patent dispute.
He later started a dot com that is still very successful.
I hope we create whalegemma (similar to dolphingemma) so we can explain to them how to co-exist better with humans (e.g. avoid this area during their whale hunting season, travel to this area if you get sick or tangled in rope).
I've noticed 2 very obvious trends. The high-performers who previously wrote the most PRs are now using AI the most aggressively and the number PRs they produce has almost doubled. The other trend is that large teams are being split into smaller teams that work on new products. I hope the trend is more projects and less grunt coding. (I know the ai coders are not yet L6 but I suspect they will achieve L3-L5 this year).
I suspect we need to build MCP servers that prevent destructive commands.
For example, we need a "bash" tool doesn't invoke /usr/bin executables directly.
The agent should think it is invoking a unix command but those commands are proxies that prevent destructive operations with no ability for an agent to circumvent the restrictions. If there isn't a MCP server for your specific setup/need, building one just for your need should be your first step.
I use 2 32" 4K which cost about $800 for both monitors. The small gap between the monitors is annoying but I can't really justify paying $2k more. Also there is a samsung dual 4k that is about the same price as the dell.
Moving my head to see everything doesn't bother me. I also have a setup with 3 32" 4k which I find a little too wide but in that setup 1 monitor connects to different computer.
The AI market is hard to predict due to the constant development of new algorithms that could emerge unexpectedly. Refer to Ilya's interviews and other sources for insights into the necessity of these new algorithms:
https://www.youtube.com/watch?v=aR20FWCCjAsDeepSeek is a valuable product, but its open-source nature makes it difficult to displace larger competitors. Any advancements can be quickly adopted, and in fact, it may inadvertently strengthen these companies by highlighting weaknesses in their current strategies. Despite this, DeepSeek is genuinely helpful.
There is an interesting exploration of human networks discussed in this veritasium video:
https://www.youtube.com/watch?v=CYlon2tvywABasically, when network connectivity increases, the "bad" nodes can overwhelm the "good" nodes. The other ideas discussed are really interesting; well worth watching.
If DOJ knows the destination wallets can they trace the transactions ...
into the wallets back to the victims?
from the wallets forward to the others working with the scammer?
I think those of us who want youtube without the distractions should start a campaign to convince google to make
https://www.youtube.com/learning better. I think even tweeting about it so that our efforts are reported by large news outlets would work. If there aren't enough users asking for this feature, we will be ignored. Someone needs to build a website that coordinates our efforts.
Preventing garbage just requires that you take into account the cognitive limits of the agent. For example ...
1) Don't ask for large / complex change. Ask for a plan but ask it to implement the plan in small steps and ask the model to test each step before starting the next.
2) For really complex steps, ask the model to write code to visualize the problem and solution.
3) If the model fails on a given step, ask it to add logging to the code, save the logs, run the tests and the review the logs to determine what went wrong. Do this repeatedly until the step works well.
4) Ask the model to look at your existing code and determine how it was designed to implement a task. Some times the model will put all of the changes in one file but your code has a cleaner design the model doesn't take into account.
I've seen other people blog about their tricks and tips. I do still see garbage results but not as high as 95%.
The Anthropic's research on how LLMs reason shows that LLMs are quite flawed.
I wonder if we can use an LLM to deeply analyze and fix the flaws.
What do you think of the recent anthropic research on how LLMs reason? It is clear from their shallow analysis that LLMs have very serious reasoning weaknesses. I wonder if those weaknesses can be "addressed" if we build LLMs that can do deeper analysis and use RL to self-improve. LLMs improving LLMs would be a very impressive step towards AGI.
I'm hoping this alternative image prompt preprocessing technique gets more attention:
https://art-msra.github.io/
Basically, the user's image prompt is converted to a several prompts to generate parts of the final image in layers which are combined. The layers are still available so that edits can cleanly update one section without affecting the others.
I saw an interesting quote from the deekseek CEO:
"When tackling a task, experienced people might reflexively tell you 'this is how it should be done,' while those without experience will experiment repeatedly, think deeply about the best approach, and ultimately find a solution tailored to the actual circumstances."
I do not think that OpenAI is out of ideas, but I do believe that large businesses have difficulty overcoming the innovator's dilemma. In addition to big businesses, there are also many small and very smart companies that are using investors' over exuberance to their advantage. Nevertheless, there are enough people trying that some are likely to find a way to develop higher-quality models at a lower cost.