suckminimally.com
I'm here because the billionaires aren't. I write code, make user interfaces, drink coffee, read, and try to make things a little better for people.
(he/him)
Known as "monitron" in some places.
http://suckminimally.com/
290 posts
224 followers
462 following
Regular Contributor
Active Commenter
comment in response to
post
Specifically this stuff is elite. Even the diet version isn’t bad!
comment in response to
post
I don't think we're connecting here. There's no such thing as a working LLM that is "trained on one thing." To read and write English, they get trained on *everything*. An LLM that is fine tuned on one task will get better at it. But your accuracy claims are unsubstantiated and, I think, doubtful.
comment in response to
post
Further, the interpretation and enforcement of these documents is governed by ever-changing laws and precedent, and their verbiage has baggage from legal tradition. Even lawyers may need to do research to come to a correct answer on this type of question. It's almost a worst-case scenario for LLMs!
comment in response to
post
The LLM you're using for this IS generic. Even if fine-tuned, the base model is trained on extensive "world knowledge" so that it can "make sense" of the user's natural language question, "read" the contract/policy, and produce a meaningful answer. A general NLP problem if ever there was one. (2/3)
comment in response to
post
I think you're mistaken. Specialized neural networks can be very accurate. For instance we use NNs for OCR and they do a great job. But the tasks you were talking about (answering questions about company policy, contract details) are the opposite of specialized. (1/3)
comment in response to
post
What is the material difference? Any LLM produces probabilistic output based on its training. Are you saying the latter is somehow not liable to get things wrong in the same way that ChatGPT is?
comment in response to
post
That’s neat, but it doesn’t seem like it’s very good at being a TV
comment in response to
post
www.youtube.com/watch?v=W_CP...
comment in response to
post
Toad must have followed the advice on the cover of that one Blink-182 album.
comment in response to
post
Hell yeah, why not?!
comment in response to
post
Unfortunately my attempts to devise a proof-of-can't-work scheme to authenticate users aren't going too well.
comment in response to
post
Hey very cool, congratulations!! 🚢
comment in response to
post
“Watch your step as you axit.”
comment in response to
post
Yes! Thank you. I am having so much trouble trying to get people to understand my concerns about the higher-order consequences of gen AI use, and you stated one of the biggest ones very well.
comment in response to
post
Thanks for the heads up. Just deleted their app and unsubscribed from their emails. Good thing there are about a dozen better burrito joints near me. I feel bad for folks in tex-mex food deserts.
comment in response to
post
“Heck of a job, Trumpie”
comment in response to
post
Oh, and thank you for your thoughtful reply!
comment in response to
post
…they’re the path of least resistance as well as the industry standard, leaving less room for developing the love of engineering you mentioned.
In hiring, I already have to sift through 100 people to find one who really cares about this work. I’m worried in 10 years it’ll be 1/1000 or worse.
comment in response to
post
I do hope you’re right. I think you’re a lot younger than me so maybe you have a better read on where people are going with these things.
My concern is that if the code generating tools (and that’s what I see people most excited about) become the default, learners will use them because (1/2)
comment in response to
post
honest question: LLM-assisted programming requires the user to judge the AI’s output in order to be successful, right? How do you think the next generation of programmers will develop the understanding and taste to do so, if they inevitably rely on LLM help from the moment they start learning?
comment in response to
post
Hey I just did this too. Did you use Kenji’s recipe by any chance?
comment in response to
post
I respect you a lot, but I’m sad you seem to be missing the nuance on why aspects of the rush to adopt AI sadden and concern people so much. We can’t just continue to make technology for its own sake, or for the sake of profit. We have to consider the humans, the future.
comment in response to
post
Besides everything else that others have pointed out, AI is the most powerful tool yet for those who wish to diminish objective truth, critical thinking, genuine human connection and the power of labor. All things we desperately need more of, not less.
comment in response to
post
I think maybe he’s saying he *thought* he was almost done before the catastrophic problem made itself known.
comment in response to
post
Ohmigosh he has so much personality!!
comment in response to
post
Absolutely I’d be doing that too. WTF
comment in response to
post
Wow, if *anything* should be hot-swappable, it’s Ethernet!
comment in response to
post
They are really really good at *sounding* like they know what they’re talking about, and yet if you ask them about any subject you’re an expert in, you can see right away that they don’t. Do you not see how pernicious that is?
comment in response to
post
If they ever break the Subscribed page I will simply never come back. You’ll have to become a Nebula creator!!
comment in response to
post
James Hoffmann has a great video where he tries all of them. He is so visibly exhausted by the end. The funniest part is how the “tasting notes” for more than half of them list a top note of “cereal.”
comment in response to
post
Wheeeee that sounds like a real anxiety-inducing travel day, I'm sorry. My list is a little longer but I'll be sure to add Turkish in case it comes up.
comment in response to
post
I was shocked to learn that apparently 12% of Americans have tried a GLP-1! I'm only at 21% of my weight loss goal but I'll probably stop updating regularly at this point; I figured I was posting about something relatively unusual, but it seems I'm preaching to the choir! (3/3)
comment in response to
post
My lipid numbers have improved too but I'm also on a fibrate so I can't attribute all of that to the Zepbound.
It still feels too good to be true that a drug can have so many positive effects on me. I'm waiting for the other shoe to drop. There's virtually no downside, apart from the bill. (2/3)
comment in response to
post
Oh no what happened?
comment in response to
post
This type of news is like ambrosia to me. I need more
comment in response to
post
Oh no! I was hoping it was just my manufacturer that was being dumb!!
comment in response to
post
Only if they come, in whole or in part, from another country. So like, everything. So, yes.
(Unless they’re discretionary purchases and you want to save up for whatever else might get nightmarishly expensive!)
comment in response to
post
It makes me smile to see them every time!!
comment in response to
post
You want “our” party to abandon my daughter to the monsters who want her not to exist. Take your withered excuse for a heart and get the hell out.