14 models tested on 960 analyst tasks: AI scores 97/100 as analyst, picks winners no better than random
HeyAmit_ · x · 2026-10-31
A fund benchmarked 14 models across 960 real analyst tasks and 101 rounds of "pick the winners." Result: AI excels at analysis (97/100) but performs no better than random guessing when choosing eventual winners — great at research, poor at judgment calls.
More from Venture
- DeepSeek's $15.8B round: Chinese state investors supplied 27%, up from 2% — rohanpaul_ai · 2026-10-31
- DeepSeek's $15.8B round: Chinese state investors supplied 27%, up from 2% — rohanpaul_ai · 2026-10-31
- Cursor's $60 billion valuation mocked: 'great investment' when agents do everything — mgostIH · 2026-10-31
- Reddit stock plunges 42% in 8 trading days as CEO says AI mode no longer drives traffic — SumitGup · 2026-10-31
- AI referral traffic grew 5.6x in 2 years but is still just 0.5% of visits, Growth Memo finds — SEO · 2026-10-31
- Scope launches Intents to track how AI agents discover, evaluate and drop off from your product — RaphaelDabadie · 2026-10-31