Does anyone know of any good write-ups on how to carry out this sort of task, for people who are reasonably technical (i.e., know how to code) but aren’t deep in the AI world? I feel like “customize a model based on a corpus of documents” (whether that’s “fine-tuning” or “RAG”) is a thing that everyone wants to know how to do but nobody actually explains in straightforward terms. (I pay for Gemini solely for access to NotebookLM for these purposes, but it would be nice to just be able to roll my own locally.)
Glad someone finally stopped pulling punches in describing Yarvin's wickedness and imbecility.
There's a kind of elegance and charm to this style of just heaping abuse on people who richly deserve it---Hunter S. Thompson was its greatest practitioner and this is in that tradition nicely.
> Glad someone finally stopped pulling punches in describing Yarvin's wickedness and imbecility
That's much too simplistic and mean spirited. We're not going to address the problems we're facing with personal dismissals. Yarvin, the man, isn't the problem. Some of his ideas are ridiculous and unworkable, but they're motivated by something real and unaddressed by many other commentators. The current zeitgeist, which is centered on personal attacks and dismissals is contributing to our problems, and inability to make any real progress at addressing them, by increasing factionalism and intellectual silos.
Edit: As to the addition you made to your comment, we live in a very different world than when Hunter S. Thompson was making his contributions.
Do people actually do this in things like slack? (One of the best things about being a professor in a non lab field is that I don't have to use things like slack.) This seems like open contempt for the reader.
I've never seen it, but I have a buddy who had a coworker like this. Would basically treat his slack as a manual copy-paste bridge to an LLM and it's was incredibly unhelpful because most questions were heavily context dependent.
I imagine this is the kind of thing you see at a large company where a good chunk of people are just coasting by doing nothing, Nelson Big Head style.
Yes all the time. Healthy conversations are shut down immediately with the "AI research" slop-wall.
Meetings are 55 minutes of speculating about what AI will be able to do in a few years. Then if you are lucky the last 5 minutes can be used to discuss a real issue.
There are actually really interesting tech intersections here, possibly. 8 years ago I published an article[1] that ended with some speculation about the possibility of automating organized consumer action to permit, for example, people to coordinate to force companies to drop abusive terms in contracts. The co-op model is in the same family.
Back in the day, my basic idea was basically this: the reason that consumer companies can impose unfair terms on people is because it's basically only a small amount of money to an individual, and it's costly for people to coordinate. For example, the expected cost to me of some company's arbitration clause is small, because the probability of any individual into a dispute with them is low. But in the aggregate it's bad: over a million customers (say) it's almost certain that one of those customers gets royally screwed by it. So: what if all those million customers of company X could all agree to automatically cancel their service contracts with X, if and only if enough other people also agree to do so that a simultaneous cancellation would cause real pain to the company's bottom line. And then enforce that agreement with automation. Some basic game theory suggests the objectionable clause immediately goes boom.
yes exactly this is the exact problem I want to see solved. I think high trust demand co-ops like JCCU(https://jccu.coop/) have solve this issue. I will definitely read your article.
Back when I worked at Google there was an internal page someone put up that denoted what they called "the YX problem": the observation that the XY problem, applied to a sufficiently great extent, creates an environment where more productivity is lost convincing one's interlocutor that X is in fact the correct problem to solve than would be lost by chasing X and having to later pivot to Y if that turned out to be wrong.
It's extraordinarily aggravating when it happens. I really wish it was something we talked about more.
In my experience, the ones who most fervently believe they have such “clues” are often the same ones who lack them. I can’t tell you the number of times I’ve encountered engineers who become indignant when someone tries to redirect them to safe, scalable, and operable ways to solve problems. What they often want is to have the problem solved their way because it would be less work for them or otherwise advance their own personal interests, regardless of the problems that doing so would create or the risks it would pose. They don’t want a discussion; they want a rubber stamp.
> because it would be less work for them or otherwise advance their own personal interests
I've seen plenty of people rationalize not wanting to learn a new thing with those words.
The thing is, this is a rationalization too. It may be true, it may also be false. The only certain thing is that they don't want to learn a new thing.
That said, this happens in small circles, where you know the context of the person asking the question. People assuming they know why a stranger is asking something tend to be wrong too.
I mean, yeah, if you’re selling someone a product it’s pretty fair for them to expect, or at least hope, that it will reduce their workload and advance their personal interests…
Can confirm, I had a coworker doing that a while ago and it was incredibly frustrating. After going through the whole quizzing you still end up with no answer to your actual question, as if they were more interested engaging with their own process and not what you were actually asking
There's a fine line here when dealing with customers. Sometimes it works well to answer the exact question, if you can, and follow up with "Can you tell me a bit why you're asking so I can understand a bit more about the problem?" Once you tease out a bit more about how they got there, it's often possible to offer better solutions and it never feels like you blew off their original ask.
People who really don't have a clue ignores the added context and answers the question that wasn't asked anyway, because they've answered that particular question before.
I'm a little puzzled by what this actually is supposed to be. The marketing material on this website suggests that it's meant to be used with a firm's Gemini or Claude API keys. ("A chat interface that reads your documents, cites verbatim, runs multi-step workflows, and drafts and edits contracts end-to-end. Plug in your own Claude or Gemini keys, and keep full control of the models you use.").
If that's true, how does it actually achieve anything with respect to client confidentiality or anything else? (For example, there's the claim "the assistant keeps full context across every conversation and every document." --- but isn't that a function of the model one uses, which is on Anthropic or Google? Ditto the claim "Documents never leave your perimeter. Compliance, residency, and privilege stay under your control." But this is only true if you're not piping them to Anthropic or Google...) Is this just a user interface?
It would be nice if these product webpages included an easy way to find documentation so that one could figure out what the product actually does. I can't find any obvious way to discern if it can be easily used with a local model running via ollama or something, for e.g.
These firms have enterprise relationships that dictate all of that. This is presumably just a frontend that takes the key as an input and plugs into that infrastructure.
Hard disagree. The iPad is a fantastic mac replacement for many purposes. I use the iPad Pro w/ the “magic keyboard” case for working essentially whenever I’m not physically in home or office in similar ways that I do my Mac, for two really big reasons:
(1) The (11-inch) size is fantastic: you get enough screen real estate to see what you’re reading and writing, but it still fits into an arbitrarily small bag and is light enough that you can comfortably walk around all day with it. The death of the original tiny MacBook Air was a huge fail for apple
(2) CELLULAR CONNECTIVITY FOR GOD’S SAKE CELLULAR CONNECTIVITY. Yes, you can always hotspot your phone, however, that’s still not nearly so reliable as a device with its own connectivity, some providers still limit bandwidth there, plus the last thing I need is extra battery drain on my phone when I’m already stressed about it.
TBF, if Apple ever brought back the original MacBook Air with modern specs and with a cellular chip, I would just take gigantic buckets full of money and throw them in the general direction of Cupertino until I got one, like, instantly. And there are definitely still compromises—-as an academic, I’ve been meaning to just write a command line front end to zotero and fling it onto a digital ocean server or something, because its iPad app is so godawful. But on the whole, I still reach for my iPad much much much more than my MacBook, for those two killer features.
I don't use it (or need it) myself, but when I was working for a sporting equipment manufacturer a number of years ago, every salesman had a cellular dongle for their laptop. We had to remind them they had direct ethernet connections when they were in the office.
Fun fact, I once bought a reconditionned laptop and the sim card of the previous owner was still on the slot. More interestingly I could use it to connect to the mobile network for at least 2 years without even knowing the PIN (and having reinstalled to linux).
Do people really not use it? I use my iPad cellular all the time. Constantly.
One of these days I'm going to buy one of those old MS Surfaces with cellular and stick Linux on it. But for the installation/drivers hassles I'd have already done so.
It really depends how frequently one is outside. I have had many laptops with cell chips and sim slot but never bothered to pay for a sim when I could just tether my smartphone connection, even when using the train. I usually plug the phone to the laptop if I need to charge the phone.
reply