AI News Daily
π
μ μΈκ³ AI νΈλ λ κΈ°μ¬λ₯Ό μλ μμ§ (μ΅λ 50κ°)
ginipick
ginipick
AI & ML interests
None yet
Recent Activity
updated a Space about 13 hours ago
ginipick/ai-news-daily published a Space about 13 hours ago
ginipick/ai-news-daily reacted to SeaWolf-AI's post with π about 22 hours ago
Why This Matters β David Defeats Goliath
MODEL: https://huggingface.co/FINAL-Bench/Darwin-4B-David
SPACE: https://huggingface.co/spaces/FINAL-Bench/Darwin-4B-david
We're releasing Darwin-4B-David, the first second-generation model in the Darwin Opus family. By evolving an already-evolved model, it achieves 85.0% on GPQA Diamond β surpassing its 58.6% original ancestor and even gemma-4-31B (84.3%) β with just 4.5B parameters.
Second-Generation Evolution
Most merges start from a base model and produce a single offspring. Darwin-4B-David breaks this pattern. The Father (Darwin-4B-Opus) was already evolved from gemma-4-E4B-it with Claude Opus reasoning distillation β a Gen-1 model. The Mother (DavidAU's DECKARD-Expresso-Universe) brings Unsloth deep tuning across 5 in-house datasets with thinking mode by default. Crossbreeding these two produced the first Gen-2 Darwin model.
Darwin V6's Model MRI scanned both parents across all 42 layers, assigning independent optimal ratios per layer. The Mother's creativity and Korean language hotspot (Layer 22-25, weight 0.95) was maximally absorbed, while the Father's reasoning core (Layer 30-40, weight 0.48) was preserved. This is "Merge = Evolve" applied recursively β evolution of evolution.
Benchmarks
Darwin-4B-David scores 85.0% on GPQA Diamond (+26.4%p over original 58.6%), evaluated generatively with maj@8 (8 generations per question, majority vote), Epoch AI prompt format, thinking mode enabled, 50 sampled questions. On ARC-Challenge (25-shot, loglikelihood), both score 64.93% β expected, as loglikelihood doesn't capture thinking-mode reasoning differences.
Why This Matters
gemma-4-31B (30.7B) scores 84.3%. Darwin-4B-David surpasses it at 1/7th the size β no training, no RL, just 45 minutes of MRI-guided DARE-TIES on one H100. The name "David" honors Mother creator DavidAU and evokes David vs. Goliath.
Organizations
updated a
Space about 13 hours ago
published a
Space about 13 hours ago
reactedto SeaWolf-AI's post with π about 22 hours ago
Post
1873
Why This Matters β David Defeats Goliath
MODEL: FINAL-Bench/Darwin-4B-David
SPACE: FINAL-Bench/Darwin-4B-david
We're releasing Darwin-4B-David, the first second-generation model in the Darwin Opus family. By evolving an already-evolved model, it achieves 85.0% on GPQA Diamond β surpassing its 58.6% original ancestor and even gemma-4-31B (84.3%) β with just 4.5B parameters.
Second-Generation Evolution
Most merges start from a base model and produce a single offspring. Darwin-4B-David breaks this pattern. The Father (Darwin-4B-Opus) was already evolved from gemma-4-E4B-it with Claude Opus reasoning distillation β a Gen-1 model. The Mother (DavidAU's DECKARD-Expresso-Universe) brings Unsloth deep tuning across 5 in-house datasets with thinking mode by default. Crossbreeding these two produced the first Gen-2 Darwin model.
Darwin V6's Model MRI scanned both parents across all 42 layers, assigning independent optimal ratios per layer. The Mother's creativity and Korean language hotspot (Layer 22-25, weight 0.95) was maximally absorbed, while the Father's reasoning core (Layer 30-40, weight 0.48) was preserved. This is "Merge = Evolve" applied recursively β evolution of evolution.
Benchmarks
Darwin-4B-David scores 85.0% on GPQA Diamond (+26.4%p over original 58.6%), evaluated generatively with maj@8 (8 generations per question, majority vote), Epoch AI prompt format, thinking mode enabled, 50 sampled questions. On ARC-Challenge (25-shot, loglikelihood), both score 64.93% β expected, as loglikelihood doesn't capture thinking-mode reasoning differences.
Why This Matters
gemma-4-31B (30.7B) scores 84.3%. Darwin-4B-David surpasses it at 1/7th the size β no training, no RL, just 45 minutes of MRI-guided DARE-TIES on one H100. The name "David" honors Mother creator DavidAU and evokes David vs. Goliath.
MODEL: FINAL-Bench/Darwin-4B-David
SPACE: FINAL-Bench/Darwin-4B-david
We're releasing Darwin-4B-David, the first second-generation model in the Darwin Opus family. By evolving an already-evolved model, it achieves 85.0% on GPQA Diamond β surpassing its 58.6% original ancestor and even gemma-4-31B (84.3%) β with just 4.5B parameters.
Second-Generation Evolution
Most merges start from a base model and produce a single offspring. Darwin-4B-David breaks this pattern. The Father (Darwin-4B-Opus) was already evolved from gemma-4-E4B-it with Claude Opus reasoning distillation β a Gen-1 model. The Mother (DavidAU's DECKARD-Expresso-Universe) brings Unsloth deep tuning across 5 in-house datasets with thinking mode by default. Crossbreeding these two produced the first Gen-2 Darwin model.
Darwin V6's Model MRI scanned both parents across all 42 layers, assigning independent optimal ratios per layer. The Mother's creativity and Korean language hotspot (Layer 22-25, weight 0.95) was maximally absorbed, while the Father's reasoning core (Layer 30-40, weight 0.48) was preserved. This is "Merge = Evolve" applied recursively β evolution of evolution.
Benchmarks
Darwin-4B-David scores 85.0% on GPQA Diamond (+26.4%p over original 58.6%), evaluated generatively with maj@8 (8 generations per question, majority vote), Epoch AI prompt format, thinking mode enabled, 50 sampled questions. On ARC-Challenge (25-shot, loglikelihood), both score 64.93% β expected, as loglikelihood doesn't capture thinking-mode reasoning differences.
Why This Matters
gemma-4-31B (30.7B) scores 84.3%. Darwin-4B-David surpasses it at 1/7th the size β no training, no RL, just 45 minutes of MRI-guided DARE-TIES on one H100. The name "David" honors Mother creator DavidAU and evokes David vs. Goliath.
reactedto SeaWolf-AI's post with π₯ 13 days ago
Post
4668
π World Model Bench β does your world model actually think?
FID measures realism. FVD measures smoothness. But neither tells you whether the model understood the scene.
We just released WM Bench β the first benchmark for cognitive intelligence in world models. The core question: when a beast charges from 3 meters away, does the model know to sprint β not walk? Does it respond differently to a human vs an animal? Does it remember the left corridor was blocked two steps ago?
Those are cognitive questions. No existing benchmark asks them. So we built one.
3 Pillars Β· 10 Categories Β· 100 Scenarios Β· 1,000-point scale
- π P1 Perception (25%) β Can it read the scene?
- π§ P2 Cognition (45%) β Does it predict threats, escalate emotions, utilize memory?
- π₯ P3 Embodiment (30%) β Does the body respond with the right motion?
All evaluation is via simple JSON I/O β no 3D engine, no special hardware. Any model with an API can participate.
We also built PROMETHEUS as a live reference implementation β runs in your browser on a T4, no install needed. Combines FloodDiffusion motion generation with a LLM cognitive brain (Perceive β Predict β Decide β Act). Scored 726/1000 (Grade B) on Track C β the only directly verified model so far. Submissions from other teams very welcome.
---
π Dataset β FINAL-Bench/World-Model
π Demo β FINAL-Bench/World-Model
π Leaderboard β FINAL-Bench/worldmodel-bench
π Article β https://huggingface.co/blog/FINAL-Bench/world-model
Part of the FINAL Bench Family β alongside FINAL Bench (Feb 2026). Feedback on rubrics and missing models always welcome!
FID measures realism. FVD measures smoothness. But neither tells you whether the model understood the scene.
We just released WM Bench β the first benchmark for cognitive intelligence in world models. The core question: when a beast charges from 3 meters away, does the model know to sprint β not walk? Does it respond differently to a human vs an animal? Does it remember the left corridor was blocked two steps ago?
Those are cognitive questions. No existing benchmark asks them. So we built one.
3 Pillars Β· 10 Categories Β· 100 Scenarios Β· 1,000-point scale
- π P1 Perception (25%) β Can it read the scene?
- π§ P2 Cognition (45%) β Does it predict threats, escalate emotions, utilize memory?
- π₯ P3 Embodiment (30%) β Does the body respond with the right motion?
All evaluation is via simple JSON I/O β no 3D engine, no special hardware. Any model with an API can participate.
We also built PROMETHEUS as a live reference implementation β runs in your browser on a T4, no install needed. Combines FloodDiffusion motion generation with a LLM cognitive brain (Perceive β Predict β Decide β Act). Scored 726/1000 (Grade B) on Track C β the only directly verified model so far. Submissions from other teams very welcome.
---
π Dataset β FINAL-Bench/World-Model
π Demo β FINAL-Bench/World-Model
π Leaderboard β FINAL-Bench/worldmodel-bench
π Article β https://huggingface.co/blog/FINAL-Bench/world-model
Part of the FINAL Bench Family β alongside FINAL Bench (Feb 2026). Feedback on rubrics and missing models always welcome!
updated a
Space 13 days ago
published a
Space 13 days ago
updated a
Space 19 days ago
Upload 19 files
#1 opened 20 days ago
by 
ginipick
published a
Space 20 days ago
liked a
Space 26 days ago
reactedto SeaWolf-AI's post with π about 2 months ago
Post
4311
FINAL Bench Released: The Real Bottleneck to AGI Is Self-Correction
We release FINAL Bench, the first benchmark for measuring functional metacognition in LLMs β the ability to detect and correct one's own reasoning errors. Every existing benchmark measures final-answer accuracy. None measures whether AI knows it is wrong.
Dataset: [FINAL-Bench/Metacognitive]( FINAL-Bench/Metacognitive) | 100 Tasks | 15 Domains | 8 TICOS Types | Apache 2.0
Leaderboard: FINAL-Bench/Leaderboard
Article: https://huggingface.co/blog/FINAL-Bench/metacognitive
Core Innovation
Our 5-axis rubric separates what no prior benchmark could: MA (Metacognitive Accuracy) β the ability to say "I might be wrong", and ER (Error Recovery) β the ability to actually fix it. This maps directly to the monitoring-control model of Nelson & Narens (1990) in cognitive psychology.
Three Findings Across 9 SOTA Models
We evaluated GPT-5.2, Claude Opus 4.6, Gemini 3 Pro, DeepSeek-V3.2, Kimi K2.5, and others across 100 expert-level tasks:
1. ER Dominance. 94.8% of MetaCog gain comes from Error Recovery alone. The bottleneck to AGI is not knowledge or reasoning β it is self-correction.
2. Declarative-Procedural Gap. All 9 models can verbalize uncertainty (MA = 0.694) but cannot act on it (ER = 0.302). They sound humble but fail to self-correct β the most dangerous AI safety profile.
3. Difficulty Effect. Harder tasks benefit dramatically more from metacognition (Pearson r = -0.777, p < 0.001).
Paper: FINAL Bench: Measuring Functional Metacognitive Reasoning in LLMs
FINAL Bench is the first tool to tell apart what AI truly knows from what it merely pretends to know.
We release FINAL Bench, the first benchmark for measuring functional metacognition in LLMs β the ability to detect and correct one's own reasoning errors. Every existing benchmark measures final-answer accuracy. None measures whether AI knows it is wrong.
Dataset: [FINAL-Bench/Metacognitive]( FINAL-Bench/Metacognitive) | 100 Tasks | 15 Domains | 8 TICOS Types | Apache 2.0
Leaderboard: FINAL-Bench/Leaderboard
Article: https://huggingface.co/blog/FINAL-Bench/metacognitive
Core Innovation
Our 5-axis rubric separates what no prior benchmark could: MA (Metacognitive Accuracy) β the ability to say "I might be wrong", and ER (Error Recovery) β the ability to actually fix it. This maps directly to the monitoring-control model of Nelson & Narens (1990) in cognitive psychology.
Three Findings Across 9 SOTA Models
We evaluated GPT-5.2, Claude Opus 4.6, Gemini 3 Pro, DeepSeek-V3.2, Kimi K2.5, and others across 100 expert-level tasks:
1. ER Dominance. 94.8% of MetaCog gain comes from Error Recovery alone. The bottleneck to AGI is not knowledge or reasoning β it is self-correction.
2. Declarative-Procedural Gap. All 9 models can verbalize uncertainty (MA = 0.694) but cannot act on it (ER = 0.302). They sound humble but fail to self-correct β the most dangerous AI safety profile.
3. Difficulty Effect. Harder tasks benefit dramatically more from metacognition (Pearson r = -0.777, p < 0.001).
from datasets import load_dataset
dataset = load_dataset("FINAL-Bench/Metacognitive", split="train")Paper: FINAL Bench: Measuring Functional Metacognitive Reasoning in LLMs
FINAL Bench is the first tool to tell apart what AI truly knows from what it merely pretends to know.
- 6 replies
updated a
Space 3 months ago
published a
Space 3 months ago
updated a
Space 3 months ago
published a
Space 3 months ago
updated 2
Spaces 3 months ago
published a
Space 3 months ago
updated a
Space 4 months ago
published a
Space 4 months ago