nightshift1

I wonder if they mentioned to those models what was the original prompt.

The whole page reads like it was vibe-written by an AI. If I'd built something as disruptive as this claims to be, I'd have spent at least fifteen minutes writing the announcement myself. Every time I see 'we' in an announcement like this, I picture one guy alone in his basement.

I think that providing modern monitors that make sense is probably the easiest way to please employees. No matter how excellent the salary is, if they are stuck working on old, small, blurry, faded screens, they'll be unhappy. It also gives a bad impression of the company as a whole. I did L1 tech support a long time ago, and people were overwhelmingly happy to see me coming to replace their old CRTs with LCDs.

The slide at 14:06 say:

By june 11: Highly persistent experimental, internal-only model begins training.

I am not sure what that means. Are they preserving notes/memories and context between runs?

Maybe they are doing it on purpose to demonstrate what it looks like when untrustworthy parties have access to frontier llm and that its development should be regulated to protect all of us. /tinfoil

It ought to be extremely easy now to use LLMs to transpile cobol and do the necessary reverse engineering to migrate off mainframes and aix servers. I think IBM will lose many customers.

but are you willing to spend the time or tokens to build it again for yourself ?

Interesting. I will try some of of those instructions but I don't mind much about the tics.

My actual complaint: end-of-turn summaries that are undecipherable unless you watched the session closely. Both Opus and Fable regularly ends hours of work with status updates phrased in crystal clear terms like:

  "**V2** debug-helper gate (R4d) | Medium-risk | Gate on `?debug=1` URL param at 3 fixture sites — **not** a localStorage flag (`browser_page` clears it every test). 284 debug-API calls ride the fixture, so R4d goes last+alone"

I think it is because the plans are crazy expensive. when you pay 200-300$/month for an online subscription, its normal to want to extract as much value as possible.