6 ms·
ive found degraded performance on models larger than 4.7. i assume its model damage from overly self righteous post training resulting in false/feigned balance
by carterschonwald 1mo ago
ive found degraded performance on models larger than 4.7. i assume its model damage from overly self righteous post training resulting in false/feigned balance imported into any long running complex task.
wish i was joking.
- retr0rocket 1mo agoAsk it about maxwellhill lmao
- brcmthrowaway 1mo agoAren't the model weights frozen?
- mceachen 1mo agoModel competence is an interaction of weights, system prompt, and harness.
- actsasbuffoon 1mo agoDon’t forget reasoning effort. We get labels like “low,” “high,” and “max.” That doesn’t mean that the numbers associated with those don’t get remapped on the backend.
- Evidlo 1mo agoI think there are other knobs that can be turned without retraining.
- kardianos 1mo agoI've switched off claude this week; the last week has been significantly degraded in ability, many more screw-ups.