This is the funniest one on the list to me.
Top 40 occupations with highest AI applicability score (most at risk, sorted alphabetically):
- Thread starter Brolic
- Start date
This is the funniest one on the list to me.
Artificial Intelligence
Not Allen Iverson
I can do that… more effectively… we pack and deliver like ups trucksI’m going back to peddling rock
The Pledge
If it’s a lie, then we fight on that lie.
Oh god please no
Artificial Intelligence
Not Allen Iverson
My job is safe for now.. but the ones I went to school for and studied for
they are literally working to create ai that can do just that, make novel insights.
Sama tweet on gold medal performance, also says GPT-5 soon
OpenAI researcher confirms IMO gold was achieved with pure language based reasoning
[Discussion] What are the new techniques he's talking about?
Commented on Sat Jul 19 14:02:03 2025 UTC
Let's assume OpenAI employees are being forthcoming.
Jerry Tworek: all natural language proofs, no evaluation harness, little IMO-specific work, same RL system as agent/coder
Alexander Wei: no tools or internet, ~100 mins thinking, going beyond "clear-cut, verifiable rewards," general-purpose RL + test-time compute scaling
Sheryl Hsu: no tools like lean or coding, completed the competition in 4.5 hours, the models tests different strategies/hypotheses and makes observations
What they're saying is that they've gone beyond RLVR. Which is pretty wild. With RLVR, you only get reward feedback after completing an entire task. The signal is faint. It sounds like they've figured out how to let the model reward itself for making progress by referencing an internal model of the task. Makes sense? Let the model make competing predictions about how things will unfold, and it can use these to anchor its reasoning.
Let's assume OpenAI employees are being forthcoming.
Jerry Tworek: all natural language proofs, no evaluation harness, little IMO-specific work, same RL system as agent/coder
Alexander Wei: no tools or internet, ~100 mins thinking, going beyond "clear-cut, verifiable rewards," general-purpose RL + test-time compute scaling
Sheryl Hsu: no tools like lean or coding, completed the competition in 4.5 hours, the models tests different strategies/hypotheses and makes observations
What they're saying is that they've gone beyond RLVR. Which is pretty wild. With RLVR, you only get reward feedback after completing an entire task. The signal is faint. It sounds like they've figured out how to let the model reward itself for making progress by referencing an internal model of the task. Makes sense? Let the model make competing predictions about how things will unfold, and it can use these to anchor its reasoning.
│
│
│ Commented on Sat Jul 19 16:04:06 2025 UTC
│
│ Noam and others have said RL for unverifiable rewards.
│
│ We know this is what they did. We know it's a big deal. Like that paradigm scales up to writing great novels and doing hours of low context work (as we saw in coding competition this week).
│
│ We don't know what was actually done to make that paradigm work, but this is a good guess
│
│
│ Noam and others have said RL for unverifiable rewards.
│
│ We know this is what they did. We know it's a big deal. Like that paradigm scales up to writing great novels and doing hours of low context work (as we saw in coding competition this week).
│
│ We don't know what was actually done to make that paradigm work, but this is a good guess
│
Commented on Sat Jul 19 13:17:37 2025 UTC
Since it seems DeepMind also has gold, their inevitable blogpost could give us some pointers.
Though from previous history, it always feels like the super impressive math results don't necessarily translate to other areas' capabilities just as well, so their new techniques could be very tailored to math-oriented CoT, I have no idea.
Tackling the IMO specifically was already a well-known challenge being optimized for (I assume through math formalizers), so we'll need a lot more technical detail from them to know how actually "general" their general LLM is here. (EDIT: Jerry Tworek (@MillionInt) | https://nitter.poast.org/MillionInt/status/1946551400365994077 | https://xcancel.com/MillionInt/status/1946551400365994077 | Jerry Tworek @MillionInt, Twitter Profile | TwStalker trained general models rather than optimizing specifically for the IMO. Really impressiv, damn. It's possible their new techniques still suit formal math proofs better than anything since it's a pretty valued research area since 2023, but the fact the model is actually a general reasoning LLM is seriously impressive)
From what Noam said though it's definitely related to TTC.
Since it seems DeepMind also has gold, their inevitable blogpost could give us some pointers.
Though from previous history, it always feels like the super impressive math results don't necessarily translate to other areas' capabilities just as well, so their new techniques could be very tailored to math-oriented CoT, I have no idea.
Tackling the IMO specifically was already a well-known challenge being optimized for (I assume through math formalizers), so we'll need a lot more technical detail from them to know how actually "general" their general LLM is here. (EDIT: Jerry Tworek (@MillionInt) | https://nitter.poast.org/MillionInt/status/1946551400365994077 | https://xcancel.com/MillionInt/status/1946551400365994077 | Jerry Tworek @MillionInt, Twitter Profile | TwStalker trained general models rather than optimizing specifically for the IMO. Really impressiv, damn. It's possible their new techniques still suit formal math proofs better than anything since it's a pretty valued research area since 2023, but the fact the model is actually a general reasoning LLM is seriously impressive)
From what Noam said though it's definitely related to TTC.
GPT 5 won't get Gold IMO capabilities
With new OpenAI thinking model , order of magnitude of thinking time is now in a standard work-day range.
Post too soon, and you publish a fossil
I am feeling extremely anxious over the chatgpt Math olympiad results, what exactly are humans supposed to do now?
Posted on Sat Jul 19 11:18:18 2025 UTC
/r/singularity/comments/1m3tras/i_am_feeling_extremely_anxious_over_the_chatgpt/
I loved to learn new things, and from a personal perspective, always wanted myself to be smarter than my previous self.
I loved math and physics.
Now I feel, all that is in vain, as this LLM is going to do what I want to do, and do it even better.
The other day I was making a 3 body problem visualiser for half a day. But some guy on twitter one-shotted a black hole visualiser using Grok Heavy.
I liked doing the "intellectually heavy" tasks. Now? I feel LLM will defeat me in this. If not today, 2 years from now. What exactly am I supposed to do. Art? Gone. Music? Gone. Programming, my passion? Gone. Math and Physics? Going soon. The only thing left to do is be a company founder of sorts, forming just the problem statement, and use these tools to solve problems. But I wanted to be the problem solver.
Edit : Art, music and other fun things may still be relevant. But when its about pushing the boundaries of humanity, I feel humans will no longer be needed.
/r/singularity/comments/1m3tras/i_am_feeling_extremely_anxious_over_the_chatgpt/
I loved to learn new things, and from a personal perspective, always wanted myself to be smarter than my previous self.
I loved math and physics.
Now I feel, all that is in vain, as this LLM is going to do what I want to do, and do it even better.
The other day I was making a 3 body problem visualiser for half a day. But some guy on twitter one-shotted a black hole visualiser using Grok Heavy.
I liked doing the "intellectually heavy" tasks. Now? I feel LLM will defeat me in this. If not today, 2 years from now. What exactly am I supposed to do. Art? Gone. Music? Gone. Programming, my passion? Gone. Math and Physics? Going soon. The only thing left to do is be a company founder of sorts, forming just the problem statement, and use these tools to solve problems. But I wanted to be the problem solver.
Edit : Art, music and other fun things may still be relevant. But when its about pushing the boundaries of humanity, I feel humans will no longer be needed.
Sam Altman on the model
He is starting to beleive
GPT-5 reasoning alpha
Posted on Sat Jul 19 11:40:35 2025 UTC
https://github.com/SecureBio-ai/bio...6f78ed08b880fa819e7c5/configs/config.yaml#L12
https://github.com/SecureBio-ai/bio...6f78ed08b880fa819e7c5/configs/config.yaml#L12
OpenAI achieved IMO gold with experimental reasoning model; they also will be releasing GPT-5 soon
Vandelay
Life is absurd. Lean into it.
It's gonna effect everything, only way you'll really be safe is if those on the top fukk with you.
Feudalism-for-all... yay!
Feudalism-for-all... yay!
Fillerguy
Veteran
I'm sure AI wrote this list. Jobs like Data Scientists are experiencing an uptick in work productivity, like in my job for example.
AI can't replace jobs that require a Human to interpret information to other Humans. Anyone who's using AI to make judgment decisions is setting themselves up for massive lawsuits. These LLMs will straight up ignore data and make shyt up. And if the prompt is wrong, it will lead to the wrong conclusions.
AI can't replace jobs that require a Human to interpret information to other Humans. Anyone who's using AI to make judgment decisions is setting themselves up for massive lawsuits. These LLMs will straight up ignore data and make shyt up. And if the prompt is wrong, it will lead to the wrong conclusions.
Why couldn’t ai underwrite a policy.
How the hell is AI gonna be a passenger attendant? Are they talking about robots?
Surprised that artists/graphic designers is not on that list.
I'm trying to figure out who wants to talk to AI on a customer service line? Cause I hate that shyt.
Two years they will have robots they can sling rock and have a wicked jump shotI’m going back to peddling rock
I do not see network engineers, so i good for now......
Gulp