Claims Browser

We don't actually know whether twelve common assumptions about intelligence are valid

7/27/2026

Confidence: 70%Source

general

critique

Bearish

academic

Common assumptions about intelligence may limit the scope of what we can build as intelligent machines

7/27/2026

Confidence: 80%Source

general

critique

Bearish

academic

It's possible to build impressive machines under current assumptions, but they may not lead to true understanding of intelligence

7/27/2026

Confidence: 60%Source

general

critique

Neutral

academic

Benchmark results reported as scalar numbers like '75% on XYZ' are completely meaningless without efficiency scores like cost per task

7/27/2026

Confidence: 70%Source

benchmarks

critique

Neutral

lab researcher

Francois Chollet

7/27/2026

Confidence: 85%Source

robotics

critique

Bearish

critic

Waymo driverless vehicles are ubiquitous in SF (visible within 2-3 blocks) while Tesla Robotaxis are rare and require human drivers

Existing unlearning benchmarks evaluate solely at the output level, leaving open whether unlearning truly erases knowledge or merely obfuscates it

7/27/2026

Confidence: 80%Source

safety

critique

Bearish

academic

Existing world models incorrectly entangle physical dynamics with pixel rendering and require continuous visual observation to sustain motion

7/27/2026

Confidence: 85%Source

multimodal

critique

Neutral

unknown

The current AI policy regime is still a fiasco because decisions are made ad hoc by people who don't understand how this works, rather than having a systematic regime

7/27/2026

Confidence: 70%Source

policy

critique

Bearish

independent

Zvi Mowshowitz

7/27/2026

Confidence: 85%Source

agents

critique

Neutral

academic

Existing Active Few-Shot Learning methods overlook models' internal dynamics by relying on output-level signals, missing specific knowledge gaps

Existing approaches for understanding human behavior in embodied AI rely on implicit representations and disregard structured reasoning over scene dynamics

7/27/2026

Confidence: 80%Source

multimodal

critique

Bearish

academic

Global image metrics fail to capture blurred diagnostically relevant structures or reconstruction failures in accelerated MRI

7/27/2026

Confidence: 80%Source

general

critique

Neutral

academic

Deep models routinely misclassify out-of-distribution inputs with high confidence in medical diagnostic settings

7/27/2026

Confidence: 80%Source

safety

critique

Bearish

academic

Most existing OOD detection literature assumes balanced datasets and lacks comprehensive assessment across diverse clinical OOD scenarios

7/27/2026

Confidence: 85%Source

safety

critique

Bearish

academic

Transformer-based forecasting models typically treat all time points uniformly and may underrepresent rare extreme patterns

7/27/2026

Confidence: 80%Source

general

critique

Neutral

academic

There is limited theoretical understanding of how different distributed SSL frameworks respond to data heterogeneity

7/27/2026

Confidence: 75%Source

infrastructure

critique

Bearish

academic

Existing test generation and update benchmarks often isolate the test from the code change and rely on static metadata that doesn't verify executability

7/27/2026

Confidence: 80%Source

benchmarks

critique

Bearish

academic

Computation and Language

7/27/2026

Confidence: 85%Source

multimodal

critique

Bearish

academic

Existing cloud removal approaches for optical remote sensing prioritize visual realism while overlooking impact on downstream analytical tasks, leading to semantic drift and degraded performance

Standard MLLMs struggle to effectively model panoramic properties like severe polar distortion and continuous cylindrical topologies, significantly degrading target detection accuracy in 360-degree environments

7/27/2026

Confidence: 80%Source

multimodal

critique

Bearish

academic