pull down to refresh
Amazing answer as usual optimistic thank you
I don't want to spend months to work out something that I know is good, to then just throw it out because it is obsolete by the time it's done
I'm not in tech but I imagine you've been facing this in all areas for years no?
At my fiat mines they bought a super-duper management system that took about three years to implement
The go live date kept being delayed which I assume is standard and they eventually went live with defects, only for the system to be utterly out of date with the current market offerings
And this was pre convid!
I'm not in tech but I imagine you've been facing this in all areas for years no?
Not really. There was generally no need to be this early an adopter because there was no FOMO. The pressure is insane right now and we're all using poorly engineered technology that needs massive development on the consuming side to get working. We see people losing their jobs because they are automated away by LLMs that are at best an experimental, wonky integration. I've not seen it this bad in 35+ years
The go live date kept being delayed which I assume is standard
I have done a couple of mega projects and it's always a battle between realistic timelines and attractive timelines, but a delivery delay due to quality assurance is a good delay. I'm more worried about the lack of quality assurance: no delays, poor quality.
I've not seen it this bad in 35+ years
Worrying times indeed, not much to say to that ðŸ˜
If you make AI a taboo in school, then post-school no one knows how to properly use it and it is more likely to be abused. Ideally, you'll want to enable your students for the future.
But at the same time, the technological maturity isn't there yet, at all. End-to-end, it's riddled with bugs and experiments labeled as the one and only solutions, released and hyped to the public way too early.
Example: a year ago, everyone would have told you that MCP is the holy grail, but today it's CLI + markdown (Skills). So imagine teaching AI. By the time that you have created a course on how to safely use MCP, you're rugged by the industry.
MCP and CLI tools both have downsides that probably should (but currently don't) prevent long term institutionalization. But not giving
grepto a bot right now is kind of a self-own. Worse, APIs that you'd want your bot to interface with are often hard to properly secure, and yolovibed new APIs are not going to help in that department either; but currently, that's nearly all we're getting.This is also why I have issues in trying to build the hitchhiker's guide to the AI universe: it's too fluid. I don't want to spend months to work out something that I know is good, to then just throw it out because it is obsolete by the time it's done.