I’m currently working on five projects, but they’re all blocked due to upstream technical issues or personnel issues. Perhaps layoffs and budget cuts were a bad idea.
Got time to lean, got time to clean 
But for real, yeah. My company has been very slow. We basically sell modeling expertise B2B, but right now everyone wants to do LLMs/GenAI, and we don’t have a good selling point there – why pay us instead of just hitting OpenAI’s endpoints directly? Very different situation if you need a CNN, a custom model, a classical statistical learning model, etc., that we can build from scratch. Quite frankly, we’re probably toast.
@Winslow
Maybe a hot take, but there’s gonna be a big pendulum swing back towards classic ML in the coming years.
Nash said:
@Winslow
Maybe a hot take, but there’s gonna be a big pendulum swing back towards classic ML in the coming years.
Why, is there something wrong with buck-ass wild hallucinations?
@Rory
Not if you listen to my friend who squats in an attic and does ketamine 3-4 times a week.
Winslow said:
@Rory
Not if you listen to my friend who squats in an attic and does ketamine 3-4 times a week.
Holy shit you know Elon Musk?
Winslow said:
@Rory
Not if you listen to my friend who squats in an attic and does ketamine 3-4 times a week.
Jesus Christ, that is way too much ketamine.
@Channing
Now this is data science.
Nash said:
@Winslow
Maybe a hot take, but there’s gonna be a big pendulum swing back towards classic ML in the coming years.
“Maybe a hot take, ” (Meant in good spirit btw, not trying to be needlessly mean)
I think this is right, tech is unfortunately massively prone to hype-based boom and bust cycles, and while each inevitably leaves some mark on society and how we do things, none so far have stuck around like they were at their peak or delivered anything close to that ultimate breakthrough they promise.
Edit: this is not to say LLMs/transformers/etc don’t have benefits, just a cautionary tale about getting too caught up in things and to not apply AI to problems it’s not well suited for, AI winters are a thing for a reason.
@Bailey
I think the PC, the internet, and mobile phones have definitely fulfilled their hype. Also, I would argue that the success of TikTok, Spotify, and Netflix has validated somewhat the success of recommendation systems.
@Jess
Your second example is a very good one that I never really connected in my head. Thanks.
Edit: and anecdotally, somehow Spotify figured out that I would like heavy psych just from me liking black metal and All Them Witches. That shit works, man.
@Bailey
Sure, hype cycles exist, but dismissing AI’s long-term impact is like suggesting we’ll all be commuting on horses again once the ‘car craze’ dies down.
Meade said:
@Bailey
Sure, hype cycles exist, but dismissing AI’s long-term impact is like suggesting we’ll all be commuting on horses again once the ‘car craze’ dies down.
Yep, that’s definitely what I said, mate 
Nash said:
@Winslow
Maybe a hot take, but there’s gonna be a big pendulum swing back towards classic ML in the coming years.
Yep, we seem to be just past the ‘peak of inflated expectations’ point of the Gartner hype cycle.
@Dean
Me trying to come up with place names for my vaguely mysticism and/or philosophy/introspection-themed hiking-centric DnD campaign:
- Peak of Inflated Expectations
- Trough of Disillusionment
- Slope of Enlightenment
- Plateau of Productivity
@Bailey
These could also pass for John Romero level names in the original Doom games.
@Winslow
If everything you say is true, you might as well write a quick, cheap wrapper around OpenAI and then once you have a customer, get your sales team/engineers to upsell them on something that works.
Time to write those unit tests!
Max said:
Time to write those unit tests!
But mooooom…
Are any of these based on gov’t contracts? Because it’s a new fiscal year, so a lot of contracts are written around that timeline.