After first firing half their AI staff, to follow up with reorganizing BOTH AI departments so most survivors don't trust their managers?
Oh and the OG AI department at Google had essentially everyone fired (you know, the one that had linguists) and then the AI department that took over was taken apart, half fired, to have it's corpse picked over by Deepmind. Everyone who mattered left (over 40) with only ONE real exception.
Meanwhile firing a third of the rest of the company, to make sure that whoever remains encounters company morale somewhere between mandatory fun and PIP.
Oh and you're wondering about the management reaction? They canceled PIPs (you're now fired when you'd normally have gotten a PIP)
Which also resulted in many memes of people who just don't care anymore directly criticizing leadership. Things like "Wondering about senior management? Just ask yourself how this can be made worse. For example: how can a PIP be made worse? This is how"
ha, exactly... like, the % change could be minuscule (or worse, it might only be a perceived difference, the actual quality may have regressed, or the scenario just didn't lend itself to that specific model) but people will be on here proclaiming that they're now shipping 10x the number of PRs.
for me at least, yes. just wrote it to coworkers this afternoon. Behaves way more "stable" in terms of quality and i don't have the feeling of the model getting way worse after 100k tokens of context or so.
What i notice: after 300k there's some slight quality drop, but i just make sure to compact before that threshold.
reply