mmcwilliams

13 ms·

61.
by mmcwilliams 2y ago
They're perfectly bannable if OpenAI is able to flag certain prompts and certain client IPs. Payment modes and accounts aren't the only indicators here. The real problem is building such a brittle system that it can be cut off by
62.
by mmcwilliams 2y ago
It's no skin off my back if they're violating it, just clarifying that was what you could be describing. Considering this post is about a student facing expulsion, what you were saying could lead the developers to worse outcomes.
63.
by mmcwilliams 2y ago
If the university offers an API then it's not a "scrape". If you're describing unauthorized use of an API then you've disclosed a possible CFAA violation, not web scraping.
64.
by mmcwilliams 2y ago
It seems that the preference is less about understanding or misunderstanding the technical requirements but more that it moves a capital expenditure with some recurring operational expenditure entirely into the opex column.
65.
by mmcwilliams 2y ago
You're right that they are not the actor performing the filtering but neither are the students being discriminated against the actors performing the filtering. You spell it out in your post; they are "leveraging that behavior&quot
66.
by mmcwilliams 2y ago
It definitely is and is made worse by institutions like MIT and Harvard that don't pay their full tax burden to the city due to the PILOT program. They're allowed to accrue more and more real estate while paying a fraction of the
67.
by mmcwilliams 2y ago
WER is a decent metric to compare models but there's a difference between mistranscribing "effect" for "affect" and the kind of hallucinations Whisper has. I've run thousands of hours of audio through it for co
68.
by mmcwilliams 2y ago
Do you find that the entities committing fraud are using generative AI tools to facilitate the crimes?
69.
by mmcwilliams 2y ago
Happy to beta test! I'm available at hi at username dot com
70.
by mmcwilliams 2y ago
This is pretty slick. I like that you have Umami support. It would be nice to wrap all of this up in a Dockerfile but that wouldn't be too hard to do for my own purposes. It's been a while since I've used Stripe outside of th
71.
by mmcwilliams 2y ago
Seems to be a selective rejection, as X complied with censorship orders from India [0] and Turkey [1]. [0] https://www.bloomberg.com/news/articles/2024-02-22/india-ord... [1] https://balkaninsight.
72.
by mmcwilliams 2y ago
It's obviously down to taste, but what I've seen over and over is a comment per line which to me is excessive outside it being requested of absolute beginners. That happens and also the model can't decide if it wants the comm
73.
by mmcwilliams 2y ago
Different in an exact string match but code that is copied and pasted from ChatGPT has a lot of similarities in the way that it is (over) commented. I've seen a lot of Python where the student who "authored" it cannot tell me
74.
by mmcwilliams 2y ago
It's a ~600 line file with extensive comments, probably 300 of code. It wouldn't be too hard to type this at all.
75.
by mmcwilliams 2y ago
Oh excellent! This looks really good, thank you.
76.
by mmcwilliams 2y ago
This is really cool! I've had a lot of fun using MiDaS[0] for creating stereograms and also in robotics for depth estimation. [0] https://github.com/isl-org/MiDaS
77.
by mmcwilliams 2y ago
I would assume the argument could be made that "can be run locally", "w2v can be trained faster than other models" and "doesn't cost per-call to use" to be some arguments for why someone would try this.
78.
by mmcwilliams 2y ago
I've used YOLOv5 models in robotics for object detection. While VLMs are great at describing images more generally, having a bounding box of a detection with a confidence score is very useful when paired with depth cameras for locating
79.
by mmcwilliams 2y ago
Because journalist isn't a licensed job in the state of North Carolina and is protected by the first amendment if you're asking seriously. I personally don't think it should be banned for unofficial purposes but the clarifica
80.
by mmcwilliams 2y ago
This doesn't seem to be an accurate portrayal of the ruling. From the linked Institute of Justice post[0] > Yesterday, a three-judge panel of the 4th U.S. Circuit Court of Appeals ruled that North Carolina may ban the creation of ae
81.
by mmcwilliams 2y ago
I have no special insight into what they're actually doing, but speeding up and slowing down speech have been features of SSML for a long time. If they are generating a similar markup language it's not inconceivable that it would
82.
by mmcwilliams 2y ago
This is a very selective and ahistorical definition of "strike". Strikes regularly block shipping lanes and scab workers from entering so they often prevent others from doing work on a job site.
83.
by mmcwilliams 2y ago
I might be misremembering this but I was under the impression that Ceph offered the same or very similar object storage API prior to Amazon building S3.
84.
by mmcwilliams 3y ago
Those streaming services have other revenue sources like embedded advertisements which underwrite large amounts of those production costs.
85.
by mmcwilliams 3y ago
I'm maybe naive in assuming that there is some kind of investigation going on that isn't playing out in public, but I would assume that years of emails going through Microsoft servers might make the identities of the cohorts in th
86.
by mmcwilliams 3y ago
I agree with you, I'm just unclear about the process by which people think this system could be updated for accuracy. I think manually verifying every possible answer means that the solution is to not use transformer models for this ta
87.
by mmcwilliams 3y ago
In this case, where a RAG system is pulling from city legal code, what is the fix? Handling edge cases for individual laws it gets wrong?
88.
by mmcwilliams 3y ago
Sure, very generally we're doing PEFT starting with insights from examples very much like this one [0] and have gradually built our own tooling and customized the approach a lot as the underlying Huggingface libraries have progressed e
89.
by mmcwilliams 3y ago
This is anecdata but "good enough" is relative. I've finetuned TinyLlama with the same dataset and technique as Llama2 7B for on-device purposes (not for cost or privacy but for physical hardware that have to run offline and
90.
by mmcwilliams 3y ago
It would appear to be percentage of population.

More ›