Nate Gaylinn

Ugh, LinkedIn is the worst!

I try to disable all the notifications that aren't directly relevant to me, but they keep inventing new kinds of notifications that I have to opt out of! So annoying.

Trying to do that this morning, I see they have "simplified and regrouped" their notification settings, which is hilarious, because they're showing me a list of 14 top-level notification categories, each with its own tree of sub-categories beneath it. There must be a few dozen different categories of notification, each with multiple options within.

I think I found the one I needed to turn off? I have no idea. This is actively hostile UX.

#linkedin #ux #ui

I'm so confused. I just found a small body of literature applying LLMs to reinforcement learning type tasks, exploring the use of LLMs for "autonomous decision making."

I guess people are building more LLM agent systems, and we ought to understand them and what makes them better / worse at what they do.

But I still feel like LLMs are fundamentally not suited to decision making tasks. They don't weigh options and decide. At best, you could say they interpolate what a reasonable choice might look like based on the examples of people making choices in their training data.

That's... really not the same thing! Like, not at all. It's impressive that this sometimes works, but this seems very silly to me when we could be using actual RL systems that really are making informed decisions from experience, with mathematical rigor to estimate the quality of those choices.

#ai #llm #rl

Why should I learn to code, when Claude could do it for me?

I realized recently, this is just like asking:
Why should I learn Spanish, when I could use Google Translate?

There are many reasons.

A language is vastly richer than any translation.

Becoming fluent lets you express yourself more eloquently, and in new ways.

It gives you access to new perspectives, and new tools for thought and creativity.

It helps you connect with a community of people like you, and all of the works they've produced.

It enriches you as a person.

We need to stop putting productive output over lived experience, personal growth, and self-expression. It's people that matter. We give work value, not the other way around. We need to remember how to live like that.

#llm #programming

Reinforcement Learning is AI for playing games, planning actions, and doing things in the real world. It's extremely powerful, but also a fun and interesting topic to learn. I'm teaching myself now.

I was always intimidated by all the esoteric jargon and math in RL. The classic Sutton and Barto text clears that up nicely, though. It's one of the best text books I have ever studied, and it reveals RL is much less complex than it sounds.

I think the excessive mathiness comes from the question "is this policy optimal?" which is a natural place to start, and has been fruitful for the field.

Except, in practice it's rare we can find optimal policies for practical problems. Also, our mathematical formalisms often don't fit, which is why there are so many variations based on what information is available, and what you want to focus on. This produces the tangled mess of names and notations, but really it's just the same idea from subtly different angles.

#rl #ai #programming

I got invited to update my last paper with some extra material. So, I went back and started probing deeper, and it has been a journey!

Wait, is this really doing what I thought?

If it's not, does that undermine everything I was saying? Or can both stories be true?

Is this even interesting any more? Is there a real effect, or is it just an illusion?

It is behaving systematically, though, with measurable effects. So I can identify the principles...

And I can use that to propose a better, simpler solution than the one I found through simulated evolution!

Which sorta proves that using an evolutionary algorithm here was overkill.

Except I was only able to design a better solution because of what I learned from my evolutionary experiment!

Honestly, it feels really good. This is how science is supposed to be, I think. I'm excited to add this new chapter to the paper, and get a little more progress and closure on this idea, even if it's not what I set out to do, or expected!

#academicchatter

I really enjoyed this Artificial Life paper.

Beer's lab focuses on finding ways to describe the remarkable things life does in formal terms, and build tools to visualize and analyze them. That's a really cool mission, I think!

In this case, the focus is on what makes an individual more or less "viable" or "adaptable" than another one. They did this in a really simple ALife setting based on chemical diffusion. They identified self-sustaining patterns, then tried poking at them to see how much disruption they took to break!

What's really cool here is that they started by analyzing a single stable form, then mapped out a network of all the other forms you could reach by nudging the first one. They could quantify how robust each one was to disruption, but also how plastic and persuadable they are, which regions of state space have desirable properties, and how likely agents are to "evolve" into those states by random disruption.

The visuals alone are amazing. :)

#alife #science

These days it sure feels like the tech industry is aggressively pushing all computing, hardware and software, onto a service / subscription model.

I find this particularly offensive because I know many of the executives doing this used to be engineers, who honed their craft on affordable personal computers.

They're denying the next generation the opportunity that made them successful. They're not just gatekeeping access to computers, but also to making better alternatives. They're stealing from our future.

Good AI research should tell us something about life, or it should help people. I hate seeing research about automating what people do. It's not a good goal for science or society! I was recently reminded of this by a paper applying LLMs to math.

This domain has many good questions: what do we mean when we say a person "solves math problems"? What are they actually doing? How is this like or not like what an LLM does? How might mathematicians benefit from this?

Instead, we get papers that pit an LLM against a human on a math problems dataset. This is great for claiming "AI has superhuman math abilities now!", but it's debatable whether good answers in a test-taking environment have anything to do with logic, reasoning, or creative problem solving. Instead of exploring to what extent LLMs are "really intelligent" vs. "stochastic parrots" (and perhaps the same question for humans), it reduces everything down to a number, one that hides the deeper problem and seems far more definitive than it is.
#ai #llm