@patience_cave

patience, jimmy, openai is nothing without its good research takes time in the patience cave.

 waiting in the patience cave
mazebench.com
Joined May 2024
  • Tweets9.3K
  • Following864
  • Followers7.2K
2.1KMedia
the strawberry man claims fable 5.5 is right around the corner insolent berry, no one is getting a new fable when im in town. get sat
16
4
2
149
4,528
happy Fable 5.5 is delayed day to all who celebrate it’s not launching today, this week, or this month. we are deep in stoptober, no one dares launch a SOTA model in stoptober.
29
7
3
527
31,815
no, Fable 5.5 will not launch this week all who are saying fable will launch this week said the same thing last week, but there is one chair you can trust, one chair who will never disappoint
29
3
2
268
19,526
what’s the matter? no fable 5.5? guess we’ll have to wait a little longer? i apologize that argon was launched sooner than expected, it won’t happen again
13
117
6,871
Gemini 4 Argon from @Google on ARC-AGI (Verified): - ARC-AGI-3: 97.5%, $4.8K (provider adapter harness), 49.8%, $5.8K (standard harness) - ARC-AGI-2: 91.3%, $0.32/task - ARC-AGI-1: 98.5%, $0.16/task Google DeepMind is the second frontier AI lab to score above 95% on ARC-AGI-3.
63
108
42
1,484
205,218
MazeBench vs Gemini Argon 4 There’s a new kid on the block Gemini Argon 4 scores 12% It’s a 2nd place score, just below GPT-6 Astra
22
14
3
226
13,438
more replies
Overall, it’s an incredible addition by the Gemini team, only the 4th agent to have a non trivial MazeBench run. These agents have mastered the 2D portions of MazeBench. But they continue to struggle with the 3D puzzles.
1
15
560
Try out this difficult open world 3D puzzle game for free online! mazebench.com

Maze Bench — Visual Spatial Reasoning in a 3D Open World

Maze Bench — ice-maze puzzles, player leaderboards, and a world editor.

mazebench.com
10
456
Fable 5.5 is not launching this week btw
19
5
1
167
9,612
Mazebench vs Wonder Full Muse Wonder Full scores 1% Muse Wonder scores 0% Impressive entry by Meta and friends
16
6
1
144
19,864
more replies
We also have a new MazeBench record for coding agents, the Schema harness scores 33% with GPT-6 Astra, up from 23% with Codex Excellent work by @guanningzeng and @HavenFeng congrats!
Guanning Zeng
@guanningzengOct 12
Replying to @guanningzeng
Over longer interactions, could executable programs remain useful? On MazeBench, Schema with GPT-6 Astra collects 33 gems and visits 139 rooms, exceeding the top-50 human median on both. Its program retains and reuses learned rules across context compactions.
1
20
889
Play MazeBench for free online! Gemini 4 Argon entry coming soon, we keep triggering the biology filters mazebench.com

Maze Bench — Visual Spatial Reasoning in a 3D Open World

Maze Bench — ice-maze puzzles, player leaderboards, and a world editor.

mazebench.com
1
12
672
another week of waiting in your favorite cave no, Gemini 4 Argon won’t launch this week GPT-6.1 Astra is not remotely possible Fable 5.5 does not exist and never will hope you have a splendid stoptober
19
115
4,379
Claude Haiku 5.5 scores 1% on MazeBench It is the first model of its class to have a nonzero score, an excellent conclusion to the Claude 5.5 series
23
9
2
560
24,624
The jump from Haiku 4.5 to Haiku 5.5 was significant, enormous value is packed into this tiny model Normally models of this size have trouble walking, but Haiku 5.5 demonstrates a primitive understanding of tool use It leapfrogged the equivalently priced GPT-6 Luna
2
1
57
3,089
some say a treasure trove of math proofs are landing today from openai there is no trove of mathematical treasures the job of mathematician is safe and sound there is no new astra or gpt-bel model dario will never release fable 5.5 sit and wait with me
5
2
44
2,384
Load more