> until reaching AGI, whatever that word means
Maybe that’s exactly it - AGI is ability to plan well?
12,594 karma · joined April 20, 2008
Maybe that’s exactly it - AGI is ability to plan well?
Interestingly, the artifacts include not just code but alao retroactive design and retroactive “execution plan”. I feed them to another agent to validate the approach and resume work with fresh context and update mission statement.
Unexpectedly, design doc and an execution plan are now outcomes of coding process, not inputs to it. I’m not sure how to think about this.
Like, lower the volume (“turn to tge side”) and say, hey Richard, can you compute 123^45 for me? Or “can you check flights”?
Also threat actors being on the inside breaking out is a bit new.
A while back a knowledgeable man was a philosopher, and they could all easily talk to each other and understand. As time went we split up - we have philosophy doctors in various fields who have very little to talk about.
This happened again when the term “engineer” fractured (even before software engineering). Software engineering itself fractured, too.
And it’s happening again now.
> Architect Financial Technologies, founded in 2023, offers futures in compute linked to the value of chips in America. The firm has also recently launched a platform where model-makers bid to provide “inference”—the useful output produced by an AI model in response to a prompt—rather than the underlying compute capacity.
I think the idea of selling inference is solid, but I’m not convinced Wall Street is going to be doing that. Superscalers are better positioned.
Inference will likely be priced by evals, specific for each workload.
It could be more productive than being angry. Or not, but worth a try.
I’m not sure if the job title tho. Software restoration?
The bill may create an incentive for Chinese labs to nerf their open weight models before release. If that works we get another 6 months until unnerfed models leak to public.
Thats what the bill gives us - a few months months to come up with the next bandaid.
What we have now is that for government to act the model has to cause enough damage to cause Congress to pass a new law to try to stop the model after it’s been already released.
Have you given much thought to diffing diagrams? The source can be version controlled and a source diff can be seen, but a visual diff would be amazing and easily sway me from writing prose to writing diagrams.
However 99% of businesses will not be able to effectively handle security reports in reasonable time. They did not need that level of competence or funding until now, and therefore they did not acquire it. Most companies that use locks don’t employ locksmiths.
Looking another way - the entire economy got suddenly hit with entire pile of technical/ security debt it has accumulated over 50 years, and only has one year to deal with it. No amount of diagnostic will fix this, this is economically infeasible.
One thing that is feasible is a government bailout for technical debt. All the engineers that were freed up from AI-induced restructuring should be employed to pay down technical debt across the entire economy. Simply updating all software to use latest libraries is a great start. Using AI to diagnose, repair, re-deploy all old software. Using AI to rewrite all old software that cannot be recompiled.
Nuanced position will not be middle ground. Single-dimensional thinking is exactly the problem that causes the argument. There are multiple reasoning modalities, each suitable to a certain type of human/agent cooperation, and there are different tasks where different modalities are applicable at different stages. There is no way to reduce this to a single axis.
Your model is outsourced labor, moving from micromanagement to co-worker.
My model is a genius intern, who needs tools to exceed me in some areas and lean on me in others.
The authors model is wielding a hammer and moving to a press.
The diversity is shaped by a dozen factors - we solve different problems or even types of problems, under different constraints, amount of time invested so far, past experience and so on.
We’re not taking face to face, so forgive me if I guess wrongly… I guess your frustration comes from trying to coax wide variety of behaviors into a single framework. I’d given up on that, so I just try to pay attention. To results people achieve to reset my expectations of what is possible. Naturally I’m heavily biased by my mental model, and it has to be that way or else I will go insane.
I am looking for agent-world feature flags, but not in a post on source control. Feature flags belong to wider SDLC.
Personally I can easily imagine one or two orders of magnitude more than that. They crossed the rubicon of not gating PRs on tests. In that land you can just roll out changes to production, a small subset of it, during code review to assess behavior. Or during key moments of coding.
I’ve been thinking a lot about this recently: what can we get if every change goes to production in 100ms, and what do we need to do to maintain stability?
Yes . And also they optimized for orders of magnitude more throughput and less latency, and fixing errors after the merge. This is new - they see enough benefit in velocity that justifies merging bugs and dealing with it later.
And it reminds me of when git came out and Linus said that git being super fast locally created a qualitative change in engineers head - it became a new way of working back then.
Sure. The bulk of your post reads as sharp criticism of “AI is a tool and needs requisite skills to operate”, promoting the idea that “AGents will do the thing for you”. Suddenly the last paragraph landed with “both AI and tools need requisite skills to operate”. By the time I got to the last paragraph I already made up my mind about your message and the contrast between two positions was far too small to jolt my attention.
The key improvement you can make is to spell out juxtaposition earlier.
Reverse engineering your original argument, and with benefit of your second comment, the juxtaposition is that the skill of wielding a hammer and the skill of operating AI are on the opposite ends of control / cooperation spectrum. The distance is in fact so huge that using hammers at starting point is naive at best and clickbait at worstt.
This brings me to my second point. Both of your comments are emotionally charged with incredulity. This did not put me, or you, into the mode of curious thinking.
As to the substance. If we were to stick with the frame of control / cooperation I would take a position closer to yours. The author is not clueless, they are early in their “journey” from left to right. We all are early, meaning that no matter where you are there is someone else right of you.
However, I found that frame far too constrictive yand left it behind about a month ago. I’d be happy to elaborate if you’re still reading. :)
There’s another pov where I am providing agent with tools/setup/etc to amplify its abilities, and that does require skill.
However the origins of theory explored the outcome of disastrous and irreparable outcome of space shuttle exploding, lives lost, and billion of dollars destroyed.
When LLM goes off the rails the cost is much lower, and retry is on the table. In fact our core competency shifted from writing code to creating environment where agents can retry cheaply and course correct based on feedback.
A modest increase in the number of turns it takes to reach decorehenve could make a difference between Decoherence and reaching the a grounded checkpoint.
A human has painful memories of past mistakes and knows which of their own thinking process needs to be double checked and which can be accepted on faith. An llm has no such memory, it can either verify every single word or none of them. And verification itself needs to be verified. So they have to stop somewhere.
This is why llm usually produce text that’s too shallow or too detailed.
When you ask the agent to double check something you gave it the missing piece - intuition about which part needs to be checked.
Did you need this to get your point across?