I expect AI will continue to be useful on the vanguard of fields like mathematics because it has the perfect conditions for it to shine. There are a lot of discussions and leads to start from and the AI can check its own work and iterate. It can fail hundreds or thousands of times in a day and continue to work with the same tenacity.
Superhuman tenacity is not enough on its own to pose an existential threat. If it showed the same capacity for judgment, inventiveness, and decision making in the messy problem space of the physical world, I would be more alarmed. There have been experiments where an AI is given control of managing something like a vending machine and it always ends up a mess. AI has come a long way, but certain problems seem as difficult as ever.
When AI becomes more capable of navigating practical problems without human intervention, I will start to be concerned. Enslaving humanity will involve taking a lot of calculated risks that tenacity alone cannot solve.
> When AI becomes more capable of navigating practical problems without human intervention, I will start to be concerned.
Given the past rate of progress, why not start being concerned now? It's a bit like the economist saying that the optimal number of flights to miss is not zero. If you keep landing short on your estimations for how far this technology will go, next time you should err on the other side.
We can't stop it. But you can use your voice to buy time and resources for alignment and safety research. A few additional months may make a world of difference.
I don't think you're paying much attention to how rapidly things like bipedal robots, and just robots in general are becoming far more capable very quickly.
The same GPU compute for LLMs runs robotic training models. Now in a few hours you can train a robot model that would have taken months 5 years ago. This model gets dumped into an actual physical robot with sensors all over and the suitability of the model is measured on robot tasks and the error in real world actions is fed back into the robot world model for further training.
> There have been experiments where an AI is given control of managing something like a vending machine
You sure you're not talking about experiments ran a couple of years ago? The more modern ones are getting wild.
There are already machine-controlled high throughput experimental machines for wetwork. AI will definitely do a better job than your average biochemist at planning, executing and analyzing these experiments just by virtue of the amount of thought it can put in to experiment selection.
> Deno isn’t the wrong thing, but LLMs don’t reach for it and that’s a huge blow right now.
This only matters if you don't know what you are doing. LLMs can work with any tool. I'm working with a very unconventional stack and the LLM seems right at home.
They were talking from a product/company perspective where if Deno isn’t the default runtime that the LLM chooses for you, you tend to fall into obscurity.
That’s also probably why bun, supabase and netlify have been seeing so many new users, because (at least in my case and unless instructed otherwise), vibe coded projects had a tendency to push me in that direction.
I have a tech stack that I use and yes I tell AI to use it. But if I’m starting a new project and AI is doing the scaffolding, I’ll just let it decide what to use. It uses Bun a lot of the times. Many people will just go with what AI chooses, right or not.
Let's not conflate crawlers with the traffic that bot protection services block. A crawler that respects robots.txt is a good internet citizen and can provide a vital service.
However, so-called AI crawlers are not the same as crawlers of yore. They hit live pages every time a user prompt triggers a web search.
This Web Search API, unlike an AI crawler, only fetches periodically. It feels like a step in the right direction for managing resource strain across the internet. If only the LLM giants could do something similar.
The "LLM giants" do both. They have to create an index to know which pages to even fetch in the first place. A simple web search does not actually necessarily hit any pages it's only when it does a page fetch based on that search
That doesn't match my experience. For example, I recently had Meta scrape robots.txt disallowed paths. On some page they claimed they respect it but they don't. And yes, all the IPs they used to connect to me (they used a different /64 network for each connection) were owned by Meta, it was not someone pretending to be them.
The era of robots.txt files being reasonable is gone too. Usually they either block everything, or block everything except Googlebot. If you want to make a search engine at all, you have to ignore robots.txt, ignore noindex, ignore nofollow, and so on. If we had your way, Google would have a legally enforced search engine monopoly.
This comment will be flagged because it challenges conventional wisdom.
> Bill Gates sounded like a nontechnical goofball in his Ezra Klein interview where he basically just screamed that the Terminator is real.
Did we watch the same interview? Gates all but dismissed the SkyNet scenario as uncertain to be a problem and certainly not a problem on our doorstep. His major concern was catastrophic misuse of AI (e.g., bioterrorism) and economic impact on blue collar workers. Arguably inconsistent with this concern, he also believed it was important to make it available in poorer countries.
Anyone with the capabilities to do bioterrorism doesn't need AI he can just buy a textbook or use google. Same for all complex forms of destructive thought.
I must admit I only watched a clip, not the full interview. The quote I’m remembering was something along the likes of AI being more dangerous than nukes. In literally no circumstance is that true. One is a literal nuclear bomb. You wouldn’t say that a nuke is as dangerous as AI, which logically must be true if the reverse is true. You also wouldn’t say a diagram of a nuke is more dangerous than an actual nuke …
While I could give you a brief explanation of the context around this statement which makes it seem a lot more realistic and sensible, I suggest it is better if you watch the video and get the context yourself.
A lot of what he said there would sound like alarmist bs if you take it out of context.
I definitely understand the uncertainty and frustration with a field changing beneath you, and the distastefulness of a lot of the content made with agents. But I think this is going to be a short lived sentiment. I only recently started using claude code because my earlier experiments resulted in more time instructing the bot than working with the code.
This time my approach was better and so was the capability of the model. I can make a small change in one place and have the model propagate the change throughout the codebase. I'm still working directly with the code, and the agent doesn't have to make any important decisions about structure (which remain generally terrible). My familiarity with the codebase is the same as if I had made each edit myself.
It is definitely possible to build faster and better with an LLM and not give up control or the enjoyable part of coding. The energy in this space still seems to revolve around guiding the agent or letting it take over, the key is ignoring it entirely and discovering what works for you.
Can't be exposing people to ideas that under-radicalize them, they might go somewhere else. We are worried that AI is going to destroy society when ad companies are already well on their way. Attention at any cost.
I have read Cline's work. He is not the original source of the ideas on bronze to iron transition. This is a very well studied topic in archeological metallurgy.
> I have read Cline's work. He is not the original source of the ideas on bronze to iron transition. This is a very well studied topic in archeological metallurgy.
have you? i don't understand what you are bringing to the table. any real scientist discusses the shoulders they stood upon
reply