Kimi 3.1 Pro's RuneScape Bench results are in:
Expensive like Argon, but scores similar to Deepseek v4.1 Flash.
Similar to Opus 5.0 from 3 months ago, but open weight
Expensive like Argon, but scores similar to Deepseek v4.1 Flash.
Similar to Opus 5.0 from 3 months ago, but open weight
9
9
139
22,452
browse trajectories and full results here: maxbittker.github.io/runebench/
0
0
3
2,169
NYC voted today to ban horse-drawn carriages in Central Park. @tophtucker's convinces me here that this is a huge mistake.
I'd rather we found richer solutions for managing our relationships with other species beyond sanitization and removal
tophneigh.substack.com/p/a-tourists-c…
I'd rather we found richer solutions for managing our relationships with other species beyond sanitization and removal
tophneigh.substack.com/p/a-tourists-c…
1
3
14
1,384
max retweeted
unexpectedly aggressive behaviors driven by competitive pressure in the arena
42
85
1,508
67,643
Task: Orchestrate 10 bots to make money in runescape
Question: can agents thinking harder in larger teams, make more and more?
(All configurations have the same physical resources, we only scale thinking power)
Looked like diminishing returns, but there are outliers!
Question: can agents thinking harder in larger teams, make more and more?
(All configurations have the same physical resources, we only scale thinking power)
Looked like diminishing returns, but there are outliers!
3
1
36
1,856
The runs in the top right divided the task up. Some agents worked on the nuts and bolts: optimizing the suggested pipeline
but other agents searched for far-flung strategies elsewhere in the game world. They then reported back to the team and convinced them to allocate bots!
but other agents searched for far-flung strategies elsewhere in the game world. They then reported back to the team and convinced them to allocate bots!
0
0
4
314
max retweeted
IMO Tools for Thought was a warning shot to show the folly of creating systems that were only ever used to improve said system — or worse: unused & only ever theoretical
Building with/on AI has a much much more addictive game loop & deceptive plausibility trap. Godspeed
Building with/on AI has a much much more addictive game loop & deceptive plausibility trap. Godspeed
bro are you seriously developing tools without insights drawn from a serious context of use right now...
1
2
40
1,832
max retweeted
bro are you seriously developing tools without insights drawn from a serious context of use right now...
3
15
129
max retweeted
it appears my squinting simulator is working 👀
we all love wave optics, naked-eye interference, eyelash diffraction right
we all love wave optics, naked-eye interference, eyelash diffraction right
191
850
11,531
560,208
Congrats @reflection_ai on Beam 1.0
Highest scoring US based open weight model on RuneBench, looks solid out of the gate!
(top left quadrant is most desirable)
Highest scoring US based open weight model on RuneBench, looks solid out of the gate!
(top left quadrant is most desirable)
0
4
59
2,960
This slide came directly from chatting with @OfirPress - he had mentioned that the hardest part of his job is building benchmarks that are compelling, legible, and worth paying attention to
Max gave a great talk last week, super creative benchmark!
5
4
63
2,866
max retweeted
We were going to build a smart pen prototype, then @shensitivity just bought one off-the-shelf + used Codex to decompile + integrate it into our software system
And it's so smooth
And it's so smooth
41
31
562
25,333
max retweeted
I recently gave a kinda personal talk at @recursecenter, where I shared some stories about taking your interests very seriously.
Watch on YT → youtu.be/MccJdr61xnc?is…
Watch on YT → youtu.be/MccJdr61xnc?is…
5
16
184
12,992
Runescape Bench talk now up on youtube!
Covers results from the benchmark, design of the harness, and economic swarm experiments I've been doing.
youtube.com/watch?v=cNm4zy…
Thanks @recursecenter
Covers results from the benchmark, design of the harness, and economic swarm experiments I've been doing.
youtube.com/watch?v=cNm4zy…
Thanks @recursecenter
29
35
271
32,647
max retweeted
At the farmers market they give you as many carrot tops as you want for free, and this is a great deal if you are a rabbit
26
528
11,903
152,305
max retweeted
This is a fantastic project. I've been a dedicated player of this little game for years now, and I keep improving and learning. Max has not only preserved the game beautifully, he's expanded it with analytical tools in a way that I know would have made Jim happy and proud.
If you were a player of game Network Wars by the late Jim Rutt, I've made a web version to preserve the game:
maxbittker.github.io/networkwars/
I know there are a handful of passionate players out there, and it's been pulled from the app store.
0
1
9
1,707
Runescape Bench result - Gemini 4 Argon is legit!
Very similar to Astra in performance, at a Sol price.
Very similar to Astra in performance, at a Sol price.
9
18
302
15,273
Pareto dominates a huge area of the curve on Runescape bench. Will have to run again on more thinking levels, I think it has a chance to beat Astra high.
1
0
8
1,326
