caseyy

42 ms·

571.
by caseyy 2y ago
You seem to misunderstand the role of LUTs in color grading. LUTs (Look-Up Tables — a simple multi-dimensional array data structure) describe how one color is transformed into another, and they are used for efficient color transformations.
572.
by caseyy 2y ago
How is color grading applied if not with LUTs? You know that when you do color grading with apps like Resolve, it is stored in memory as a LUT, right?
573.
by caseyy 2y ago
Yes, it does become complicated when the management are the laziest. The average worker will do what they are asked to for information security.
574.
by caseyy 2y ago
I'm not sure it's even that more inconvenient to do what I suggested. 2FA is probably the biggest inconvenience, but the 2nd factor doesn't have to be TOTP, it can be the employee badge, or biometrics (fingerprint). If it&#x2
575.
by caseyy 2y ago
In many of the companies I worked for, security did come cheap — enabling 2FA on SSO, not collecting data they didn't need, invalidating employee session tokens every 24 hours instead of 90 days, and so on. I am talking about cheap a
576.
by caseyy 2y ago
I sometimes scan over 300 documents at a time. They are OCR-ed, but this does not make them easy to search when they are unindexed (on cloud drives, or on disk/tape for archival purposes). So it is important to name them and tag them w
577.
by caseyy 2y ago
I've written something similar in python — a script that renames documents based on their contents using Ollama. It's very easy to write things like that, I recommend anyone interested in local LLMs try a fun project such as that.
578.
by caseyy 2y ago
Why aren't the fossil fuel companies moving to renewables and nuclear themselves? Wouldn't it make better long-term business sense? They can then keep selling fossil fuels as long as they have a market to sell them in, but if the
579.
by caseyy 2y ago
LLMs cannot compete on truthful high-quality information from experts. Only on plausible babble. If that’s the only information you search for, then your perspective makes sense. We shouldn’t be using LLMs for information retrieval at all.
580.
by caseyy 2y ago
There is a mental health cost to trauma and to dealing with very distressing ideas. Even if you heal from trauma (and not everyone does), the consequences of having gone through that experience will be a part of your story. People are often
581.
by caseyy 2y ago
The “can information be harmful?” is indeed an interesting debate. I think it can. Take social media, for example. It is associated with increased depression rates, so the information people are exposed to on these networks is harmful. Anot
582.
by caseyy 2y ago
I have heard of generalization vs memorization, but the article you shared is very high quality. Thank you. I do not think that SOTA LLMs demonstrate grokking for most math problems. While I am a bit surprised to read how little training is
583.
by caseyy 2y ago
> In this section, we aim to understand the sources of GPT’s predictive ability. Oh boy... I wonder how a neural net trained with unsupervised learning has a predictive ability. I wonder where that comes from... Unfortunately, the articl
584.
by caseyy 2y ago
They may claim, but there is no such plausible deniability. Not for lawyers using AI hallucinations, not for Tesla drivers crashing into things with FSD, not for tax fraud. People are ultimately held responsible and accountable for the way
585.
by caseyy 2y ago
Meme stocks are not a strategy to make money. It only looks insane if you cannot imagine that people trade for other reasons than to make money. But these guys have other goals — proving points, protesting short-selling, speculating for fun
586.
by caseyy 2y ago
This article cherry-picks the time window to make such a claim. You can do this for many volatile stocks. GME looks like it was probably pumped and dumped (intentionally or not), but it is up 107% month on month currently. So the report cou
587.
by caseyy 2y ago
Of course it is. That is the outcome of unsupervised learning. It doesn’t have a sense of what’s true and false, what’s right and wrong. It has learned to predict the next word very well and the prediction probability distribution was later
588.
by caseyy 2y ago
Well, it could capture only content of programs that ask to be captured, while they do. Web browsers could not ask to be captured when sensitive content is on screen (perhaps the web server could communicate that in headers). So could code
589.
by caseyy 2y ago
What does that even mean though? Does it mean: - 20% of the plastic in the entire device was recycled? - 20% of the plastic in the box (as the context implies) was recycled? - 20% of the recycled plastic was ocean-bound and the device is al
590.
by caseyy 2y ago
> This is a bullshit leap in performance, made possible by a bullshit leap in bullshit. > We look forward to continuing to bullshit with you, our developer and Bullshit community, to bring bullshit to our bullshit and tools, and bulls
591.
by caseyy 2y ago
This conversation has drifted slightly off-topic. It was originally about the efficiency of producing good quality outputs akin to products or services with generative AI. I work in an industry adjacent to generative AI, so my views are ver
592.
by caseyy 2y ago
I think a lot of people drastically underestimate how much the LLMs mislead the user with plausible-sounding answers. I would say that whatever % of queries you found them not to give you correct answers, the actual percentage is materially
593.
by caseyy 2y ago
There wasn't ever a ChatGPT or Perplexity.ai moment for most technical tasks. I'm always evaluating these models to make sure that the people I interview remotely for my company don't cheat. There are certain questions the mo
594.
by caseyy 2y ago
AI isn't the most efficient way to retrieve accurate information or to generate quality content, it's the most efficient way to retrieve plausible information and generate poor quality art and code. And that's really interest
595.
by caseyy 2y ago
Interfering with the pathway that causes cell death is very interesting. Maybe more than freezing. Related sci-fi: https://galactanet.com/oneoff/antihypoxiant.html
596.
by caseyy 2y ago
Profitable companies attract much more VC interest and even competition.
597.
by caseyy 2y ago
Most business founders don’t need VC money and are worse off for taking VC money. I find the mindset “my pitch deck was 2 months old so I didn’t get funding” very out of touch of business realities. It is far more likely that that type of b
598.
by caseyy 2y ago
Yes, but we all get what the author means. UDP or TCP, SMTP or HTTP — most of it is transporting low value sludge whose purpose is to exploit its consumers. It involves everything from somewhat benign forms of surveillance/profiling to
599.
by caseyy 2y ago
The internet is the largest and most elaborate monument to marketing we may ever build.
600.
by caseyy 2y ago
All the car, smartphone, insulin pump, and pacemaker owners will surely appreciate this new and exciting EM pulse. And if this device does indeed cause overheating, so will all NHS amalgam dental filling owners and people with implants. Unf

More ›