I think like most hypothetical discussions, the commenters proposing these ideas aren’t interested in practical versions of the idea with tradeoffs. They imagine a perfect version of it in their minds with no downsides that accomplishes everything they want.
The demand for professional licensure doesn’t even make sense in this context. Is professional licensing supposed to stop developers from naming their packages names that LLMs produce? Is it going to force the package repos to check that everyone has a professional license before submitting packages from the United States (or other countries with licensure)? Can it be worked around by changing your country in the drop-down box to a country that doesn’t have licensing?
The calls for software licensure never seem to take into account the global nature of the Internet and software development.
Optimally, you'd probably have seniors do some "Security Compliance Certification" and the company do it, then the product has to be approved by the certified, and if an issue arises, the certified get to be reprimanded, especially the company certification in some exponentially scaling manner so that it doesn't become the cost of doing business.
2. Computing was new and mysterious and developed faster than lawmakers could understand it, and by now it's given so much power to the top 1% that they're for all intents and purposes above the law. Cosmetology licensure is from a time when legislation still helped us.
Protectionism by a de facto trade guild was always my assumption.
There are a lot of activities where bad practitioners present significant danger to society and licensure makes sense. I never understood how cutting hair rises to that level. I'd love to know how licensure in the barber profession is anything other than a bald-faced attempt at building a moat. It seems like the market could correct for a bad practitioner in the barber space pretty easily, and with little risk to society.
sooner or later command line interfaces will require background checks and be limited to a close select group of government approved individuals, e.g. like guns in japan.
This was just after the Optus leak. Some hundreds of thousands of customers' data, down to the passport and DOB level, leaked. Again. I was going to ask him whether we, the collected IT consultants in the room, simply couldn't be trusted any more.
We've proven that we can't. I firmly believe that independent companies should no longer, by law, be able to collect my identifying information. If you must identify me, the state should provide a service. You hand off to them, they validate me, they send you a token back, I'm validated.
Sadly the microphone never made it to my corner of the room.
> Slopsquatting is a type of cybersquatting.
I feel like this is going to fall under notability eventually
I'm all for having lots of small Wikipedia articles, but the past few years they've been tending toward combining small articles together. And this is more like a dictionary entry than an encyclopedia entry.
It would be a good idea to disallow registering packages which only differ by '-'/'_'. Rust's crates.io does this, so if you register `foo-bar` you cannot register `foo_bar` anymore.
Underscore is just capital hyphen.
is how to "manually" (semi-manually) tweak the LLMs parameters so we can alter what it 'knows for sure'
is this doable yet??? or is this one of those questions whose answer is best kept behind NDAs and other such practices?
They don't 'know' anything. They are a many-dimensional matrix of the next most likely syllable given all syllables that have come before (roughly speaking).
To ask what it 'knows' is to ask why a chicken crossed the road.
Inbefore people telling me "akshually we know all about bla bla bla..." no we dont.
Put differently: GPT-4 isn’t a knowledge base, it’s a *Bayesian autocomplete* over dense vectors. That’s why it can draft Python faster than many juniors, yet fail a trivial chain-of-thought step if the token path diverges.
The trick in production is to sandwich it: retrieval (facts) LLM (fluency) rule checker (logic). Without that third guardrail, you’re betting on probability mass, not truth.
At the risk of perhaps misunderstanding or committing a category error, I wonder if there's such a thing as a category of "correct" hallucinating, distinct from things that are, in some sense, "known" via training (e.g. I read about prompting of one model showing it was able to accurately recreate most of the text of Harry Potter, so clearly it's "in there" somewhere).
An interesting upshot of that could be that models "grow" their own knowledge in an evolutionary way via hallucinations that are retained rather than pruned as part of routine filtering and training.
Though I'm sure some might suggest "hallucinating correctly" is just one of the same with ordinary b function. I wouldn't agree with that but I could at least see the argument.
ic_fly2•2h ago