Math problems with GPT-4o [video]
youtube.com
youtube.com
(For those interested in my current approach: https://robodonkey.com/)
Have GPT4o running on the background and have it type out the answers as your interviewer reads aloud the questions. Share your screen and let it find the bugs for you in real time. Never get any facts wrong as you smugly correct your interviewer about the minute details of an obscure AWS Route 53 API.
People already use AI to punch of their resumes to make themselves look more attractive.
Doing great in a FAANG interview is life changing money for people from the lower and lower-middle classes. It can bring up your entire family. The stakes are high, which means people will use every tool to have an advantage.
The interview arms race just got more crazy!
I "picked out" this problem to discuss here because I find the process of "approximating someone else's skills" an interesting endeavor without a clear solution. Do you think the current remote interviewing techniques are effective? Regardless of your answer, it looks like it's going to change dramatically. I find that to be interesting I guess :)
It will be even harder than before to land a job without friends on the inside of the company vouching for you.
It gave me the feeling of dragging people along for no real reason and just waiting for them to catch up. I mean, if the AI already knows all of this, and anyone can ask it, then what's the value of people learning it?
I get that this is a fear with AI in general (obsoleting humans). But watching this really crystallizes it somehow.
Sam Altman's intro for GPT-4o was selling all the ways it could help people accomplish tasks. But I could only think, why not just have the AI do the tasks on its own?
Someone mentioned that Altman's interviews involve him talking a good bit without providing a lot of useful info. Makes sense when you consider that his role is something of a politician.
Anyway, for now, the assist idea is somewhat plausible, but with each new version it's less so.
I think that's what it is about this video: when you watch someone being taught by AI, it flips the "assist" part on its head. Doesn't make a lot of sense that the AI has to patiently teach people to get them to a place where it can then "assist" them.
It's really good at this. But it doesn't understand, it doesn't reason. It just picks the next token like it's been trained to do.
The upshot is, unless you already know how and why the generated text is incorrect, you'll be fooled into thinking it's the answer.
There's no need for the condescension, even if you had actually understood the point.
>It uses statistics to generate...
I assume most here understand your rudimentary description of how GPT works.
My point assumes that LLMs and/or some augmented form of the same continue to improve.
If, OTOH, the best we ever do is the current generation of models that hallucinate, require babysitting, and never reach or exceed human capabilities, then my point (and >90% of the discussion around AI's future impact) are non-issues.
I suppose I assumed that was obvious.
Seriously, though, its impressive.