eupolicy.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
This Mastodon server is a friendly and respectful discussion space for people working in areas related to EU policy. When you request to create an account, please tell us something about you.

Server stats:

223
active users

#thinking

1 post1 participant0 posts today

Tried to switch an LLM flow to gemini-2.5-flash today, which came out yesterday. It seems Gemini 2.5 Flash sometimes ignores the thinking budget I set! Flow is failing because the model ignores my thinking budget of 0, starts generating thinking tokens, and immediately goes over the max_tokens limit.

Apparently this was also a Heisen-bug in the preview version: github.com/google-gemini/cookb

GitHubthinking_budget=0 does not work, · Issue #722 · google-gemini/cookbookBy miroblog

alojapan.com/1298203/japan-can Japan can’t export cool without strategic thinking #can't #cool #Export #Japan #JapanNews #JapanTopics #news #strategic #thinking Douglas Montgomery is CEO of Global Connects Media and an adjunct professor at Temple University Japan. In the world of pop culture, few stories are as remarkable — or instructive — as the rise of the “Korean wave,” or hallyu. Over the past two decades, South Korea has transformed itself from a modest exporte…

Continued thread

También:

“They are not emotionally intelligent or smart in any meaningful or recognizably human sense of the word. LLMs are impressive probability gadgets that have been fed nearly the entire internet, and produce writing not by thinking but by making statistically informed guesses about which lexical item is likely to follow another.”

“It is not easy to describe things here. It is a life of unreality. … Reflection withers. … There seems to be no place at all for real thought. … Every day I say to myself: Patience, just be patient. Don’t be discouraged, no matter what. If we do what we can, the rewarding moments are bound to return.”

#KarlJaspers, writing from Germany, in a letter to #HannahArendt, 12 March 1946.

Seeking reason for #hope, wherever I can find it @bookstodon #patience #politics #thinking

Nice blog by Gary Marcus, for me personally this is not a blow, I don't see current AI as reasoning or thinking, it still will have enormous impact without these capabilities.
Also it's fascinating that models are able to solve some difficult math problems (that are not in their trainingset ) while they fail with seemingly simple other problems.
open.substack.com/pub/garymarc
#ai #llm #reasoning #thinking

Marcus on AI · A knockout blow for LLMs?By Gary Marcus

Pithy thoughts on thinking by dustin curtis

I have been stuck. Every time I sit down to write a blog post, code a feature, or start a project, I come to the same realization: in the context of AI, what I’m doing is a waste of time. It’s horrifying. The fun has been sucked out of the process of creation because nothing I make organically can compete with what AI already produces—or soon will. All of my original thoughts feel like early drafts of better, more complete thoughts that simply haven’t yet formed inside an LLM.

Thoughts on thinking

I thought I was using AI in an incredibly positive and healthy way, as a bicycle for my mind and a way to vastly increase my thinking capacity. But LLMs are insidious–using them to explore ideas feels like work, but it’s not real work. Developing a prompt is like scrolling Netflix, and reading the output is like watching a TV show. Intellectual rigor comes from the journey: the dead ends, the uncertainty, and the internal debate. Skip that, and you might still get the insight–but you’ll have lost the infrastructure for meaningful understanding. Learning by reading LLM output is cheap. Real exercise for your mind comes from building the output yourself.

Some of these resonate. I’ve also been using LLMs since the GPT3 days and following them since the GPT2 days. I remember suggesting GPT 2 as a path forward to a friend who was trying to build a classifier for their project.

I am not as worried as dustin even if the thoughts resonate with me. The value from thinking, from mulling, from iterating, from compounding had two parts – internal satisfaction and external validation (when you broadcasted the thoughts with others). There was a time where the sheer act of broadcasting earned you an audience – because not enough people shared.

We’ve gone the other end of the spectrum now with everyone sharing most of their thoughts and it’s only getting wilder. It’s time to dig back into the internal satisfaction of the thinking. The past year was revelatory in how I understand myself and the past few weeks of output in this blog has increased primarily as a result of focusing on the internal validation and satisfaction.

To me AIs are just another tool, just like Google, the internet and it’s an incredible summarizer of the collective human thought. It’s still not a particularly tasteful one and it’s certainly not an original one (even if it might seem to original to you and others as you certainly don’t know the vast collection of human knowledge that’s available). However, it’s still a tool. It’s a useful one too.

I appreciate digging into the existentialism it sometimes brings out in people and I appreciate dustin bravely sharing his. However, stick to the internal validators, and you will also leverage to make it satisfy you.

Dustin Curtis on SvbtleThoughts on thinking • Dustin CurtisI have been stuck. Every time I sit down to write a blog post, code a feature, or start a project, I come to the same realization: in the context of AI, what I’m doing is a waste of time. It’s horrifying. The fun has been sucked out of the process... | Dustin Curtis | Designer, hacker, investor, nomad. Founder of Svbtle.
“…I am trying to think…at the edge of my own understanding…in the hopes that my own philosophical perspective may help you connect some dots that I'm not even aware of yet. In speaking at the edge of my own thinking, I'm trying to feel into the gaps that may exist in your own thinking. Not because I know something you don't, but because together we might know more.”
—Matthew Segall, Imago Machinae: Made in the Image of Our Machines
https://footnotes2plato.substack.com/p/imago-machinae-made-in-the-image
#thinking #understanding #knowing
Footnotes2Plato · Imago Machinae: Made in the Image of Our MachinesBy Matthew David Segall

On thinking and benzene

cliophate.wtf – How to think

This post resonated with me. In a tiny way, I am skeptical about it given how much I found myself agreeing with it. These are the ways I think. I will wait until you go finish reading.

Go. Do it. Now.

When you are back, you can read through the parts that just makes sense.

Disclaimer: The post argues that, by definition, LLMs cannot think. I mathematically agree with this. However, I will also be the first to admit that I am not the expert to define what thinking is.

I am very interested to know and debate if we land on LLMs thinking or not. However, this post is to expand and explore how humans think. I really liked the four parts of thinking that was highlighted in the post:

Thinking in silence;
Thinking through inspiration;
Thinking by writing;
Thinking by not (actively) thinking.

I think they do a great job of highlighting the various parts of thinking as a process. However, each process is best at doing something slightly different. I break the job down as follows:

  • exposure: this is the part where you expose your brain to “the lego bricks” as posited by the author. Or the “thinking through inspiration”
  • processing: to me, this is your brain taking each lego brick and considering the various perspectives, aspects, pivots of it. This can be done by writing, by typing, in silence
  • distilling: This is the part where your brain takes the kernel / lesson of the process and then stores it, in a way to be used in another right context again.

To be clear, I believe the author has hit the various ways in which you empower and encourage thinking accurately. I just wanted to call out that there are different parts of that process that might be encouraged by each of the rituals.

Finally, your brain does something incredibly powerful. I’ve never yet seen this happen by LLMs.

This is thinking by not thinking. It is passive. It happens without you forcing it, in the subconscious, while you do other things. I don’t know why it happens. I don’t understand what processes run in our subconscious mind in the background, I only know that I’ve experienced this before.

I always think back to the (apocryphal) story of the benzene ring structure. Put me in as a believer because it’s happened to me multiple times.

Related On boredom, and the creative power behind it

cliophate.wtfHow to think - cliophate.wtf
More from Kevin Wammer | cliophate :ok:

"It’s reasonable to ask whether, as A.I. proliferates, many people won’t begin to question the value of training their own minds when computers are already so well trained. And it’s also unclear how intellectual passivity in some domains might affect our performance in others. In my life, I’ve given two best-man speeches; at the time I gave them, I was nervous (who wouldn’t be?). If A.I. had existed then, I might have at least considered asking for its help. Let’s say I resisted the temptation—but let’s also imagine that I had already employed A.I. extensively on other kinds of writing, using it to compose my e-mails and craft my presentations. Would I still have been able to write the speeches that I ended up writing? Or would my over-all capability as a writer have stagnated or declined?

It feels strange to imagine that, someday soon, we might need to start reminding ourselves to think. But that’s what artificial intelligence does—it thinks—and, in many contexts, promises to do the thinking for us. In a world saturated with technology, we already have to remind ourselves to put down our phones; to go outside; to see friends in person; to go places instead of staring at them on our screens; to have non-technological experiences, such as boredom. If we’re not careful, then our minds will do less as computers do more, and we will be diminished as a result."

newyorker.com/culture/open-que

The New Yorker · Why Even Try if You Have A.I.?By Joshua Rothman