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.

▲

kardianos 2 days ago | parent | next [-]

I've switched off claude this week; the last week has been significantly degraded in ability, many more screw-ups.

▲

brcmthrowaway 2 days ago | parent | prev | next [-]

Aren't the model weights frozen?

▲

mceachen 2 days ago | parent | next [-]

Model competence is an interaction of weights, system prompt, and harness.

	▲	actsasbuffoon 2 days ago \| parent [-]
		Don’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 2 days ago | parent | prev [-]

I think there are other knobs that can be turned without retraining.

▲

retr0rocket 2 days ago | parent | prev [-]

Ask it about maxwellhill lmao