Stephen Oman (@stephen_oman) header
Stephen Oman (@stephen_oman) profile picture

AI & ML. Doing Data Engineering @travelport. Maintaining @NetHack_LE. Writing my own code. 🇪🇺

Dublin, Ireland, EU
Joined October 2009
Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)Photo by Stephen Oman (@stephen_oman)
Happy to announce the latest release of @NetHack_LE (version 1.2.0). You can now use the seed function to make the dungeon layout reproducible across training episodes. The in-level interaction and combat is still randomly determined and doesn't impact lower level layouts.
1
3
2
27
7,780
Stephen Oman retweeted
No, "anthropomorphizing" is a technical problem, about how AI works, not something like woke language policing. We say it's wrong because it's "technically" wrong not "morally" wrong. If you fail to set the brakes when parking on a hill, and your car then rolls down the hill, you don't claim that the car "decided to drive away on its own". Such anthropomorphizing distorts what really happened -- that you failed to set the brakes. The same applies to AI. When you "anthropomorphize" the AI you fail at understanding what actually happened.
The whole "don't anthropomorphize AI" thing is very Woke 1.0. People trying to police language that is perfectly natural in some cases and ending up in a web of contradictions.
83
383
22
2,558
98,225
Stephen Oman retweeted
Aaaaaaaand it's apparently solved 🫣😅
The NetHack Learning Environment (@NetHack_LE) has been around since 2020. Since 2024, balrogai.com is evaluating frontier LLM performance on it. We are slowly seeing progress on it, but it still seems far from being solved.
19
42
5
536
90,589
Stephen Oman retweeted
It's important to learn math not because you may need to do long division as an adult, but because it promotes rigorous thinking. IMO, history and causal inference are among the most important subjects a young person can study — math being central to the second. A significant problem in public discourse right now is people's inability to critically evaluate causal claims (e.g., vaccines cause autism, deporting immigrants will solve the housing crisis, etc.). The less able people are to critically evaluate causal claims, the worse our democracy functions because people can be preyed upon for votes. You should promote anything that improves rigorous thinking, even if it doesn't have market value. In fact, we should stop thinking of education only in the context of what improves the economy. IMO, this is distinct from knowing your own address (which Huang can avoid because he's wealthy). Memorizing facts like this may or may not be important, but it's distinct from rigorous thinking.
HUANG: “.. Try getting a kid to do long division right now. Multiplication tables are starting to be forgotten. Doing square roots? My goodness. .. .. Does it matter? .. I don’t think it does.” @ezraklein @JensenHuang
196
1,657
76
11,283
504,743
(comic) Deep work
7
127
9
975
28,005
BALROG’s leaderboard has three new entries, courtesy of @creus_roger Very interesting that GPT 5.6 Sol at max effort is still within error of Gemini 3 and 3.1 pro. GPT Astra 6 on max reasoning effort however reaches new heights, and a @NetHack_LE avg. progression of 13% 🏰
6
9
2
56
13,332
Stephen Oman retweeted
So @PlayStation has no problem taking my money every year without age verification, but now that I want to contact my friends they need to see my ID? No thanks. Forcing Ireland to go through the rules imposed by the UK’s government is backwards. Nice going @PlayStation
1
1
59
Stephen Oman retweeted
...that's the problem with async
6
4
1
70
28,909
Stephen Oman retweeted
I’m convinced that a large % of programmers don’t actually like computers. As a side effect, are also perfectly happy to throw away their reasoning to a model as soon as they can. I don’t get it, at ALL. Don’t you *LIKE* understanding the magic of the machine? You do realize hand-programming (I hate that I even have to specify hand now) is fun…right?
996
715
271
8,943
971,030
Stephen Oman retweeted
The global surface temperature record is one of the most iconic figures in climate science. We now have 12 different groups of researchers from the US, UK, EU, Japan, and China (among others) that have calculated it and found nearly identical results:
170
441
41
1,180
65,014
Stephen Oman retweeted
A page called The History Of Music Channel have completely plagiarised one of my Dad's well known cartoons and are claiming it as their work. What's possibly worse is I have a feeling they used Al to rip him off!
75
587
11
10,030
166,972
Weird set of graphs - green line backs onto itself. Have these been altered from the published versions @jburnmurdoch ?
Degrowthers are trying to force developed countries to choose the environment over the economy. But in the last 20 years, we’ve had more growth and less environmental impact. Taking care of the environment is a luxury good that countries buy more of as they get richer.
3
1
10,950
Stephen Oman retweeted
Excited to co-found Recursive (@recursive_si) with an exceptional team in London and SF to create AI that experiments on how to safely improve itself, turning compute into knowledge that accumulates in an open-ended process of endless, automated scientific discoveries.
96
110
21
905
256,408
Latest release of @NetHack_LE (v1.3.0) is now available and includes rendering the dungeon with tiles for all you convoluted people out there, seeding options for getting past the pesky full moon issue and more. Thanks to all the contributors! ⚔️
1
4
1
13
2,381
Stephen Oman retweeted
Damn, I can't find my passport to turn on my computer....
58
263
39
1,769
54,119
Today is the 33rd anniversary of the day the IRA bombed Warrington town centre and killed my son Tim and 3 yr old Johnathan Ball. Words can never describe how losing a child leaves a huge hole in your heart and your family. Eternal love Tim ❤️❤️
864
2,663
94
30,835
639,716
Stephen Oman retweeted
This is more evidence that current frontier models remain completely reliant on content-level memorization, as opposed to higher-level generalizable knowledge (such as metalearning knowledge, problem-solving strategies...)
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
187
307
43
2,933
313,933
Happy St. Patrick's Day, New York.
2,964
4,072
932
29,333
4,350,567