@kida2002.bsky.social
I am not a robot, I just don't post that much
Sick! (Laudatory)
January 16, 2026 at 11:22 PM
Sick! (Laudatory)
You are probably already aware of the various operators that can direct your searches more precisely - but just in case, they may be able to help? www.reddit.com/r/everyonekn...
some youtube search parameters to help our search
www.reddit.com
January 8, 2026 at 11:53 PM
You are probably already aware of the various operators that can direct your searches more precisely - but just in case, they may be able to help? www.reddit.com/r/everyonekn...
Hi there, sorry, don't mean to bug you, but would you be willing to share examples now please? You left me quite curious! :)
January 6, 2026 at 5:32 PM
Hi there, sorry, don't mean to bug you, but would you be willing to share examples now please? You left me quite curious! :)
np, FWIW, I'm pretty sure that this method is completely out of date in the industry, and were this not just a side/hobby project by Hailey, I suspect probably something like a transformer model such as RoBERTa would likely be closer to the current state of art
January 5, 2026 at 1:14 PM
np, FWIW, I'm pretty sure that this method is completely out of date in the industry, and were this not just a side/hobby project by Hailey, I suspect probably something like a transformer model such as RoBERTa would likely be closer to the current state of art
lol, very quaint
January 5, 2026 at 1:07 PM
lol, very quaint
Interesting paper: eegilbert.org/papers/icwsm... - I was surprised that they got so much mileage out of a simple lexicon word bank with scores plus some relatively simple rules, but then saw the publication date was 2014 - a simpler time
eegilbert.org
January 5, 2026 at 12:55 PM
Interesting paper: eegilbert.org/papers/icwsm... - I was surprised that they got so much mileage out of a simple lexicon word bank with scores plus some relatively simple rules, but then saw the publication date was 2014 - a simpler time
github.com/cjhutto/vade... - from the GitHub repo linked from hatefeed.ing
GitHub - cjhutto/vaderSentiment: VADER Sentiment Analysis. VADER (Valence Aware Dictionary and sEntiment Reasoner) is a lexicon and rule-based sentiment analysis tool that is specifically attuned to s...
VADER Sentiment Analysis. VADER (Valence Aware Dictionary and sEntiment Reasoner) is a lexicon and rule-based sentiment analysis tool that is specifically attuned to sentiments expressed in social ...
github.com
January 5, 2026 at 12:45 PM
github.com/cjhutto/vade... - from the GitHub repo linked from hatefeed.ing
Oh, and are you talking about the entire loop or just specifically the coding or design focused ones please? Oh, and the SRE loop differs from the SWE one (I only ever did the latter on both sides) - specific one there you were commenting on? Genuinely curious, hadn't heard this criticism before
January 1, 2026 at 2:19 AM
Oh, and are you talking about the entire loop or just specifically the coding or design focused ones please? Oh, and the SRE loop differs from the SWE one (I only ever did the latter on both sides) - specific one there you were commenting on? Genuinely curious, hadn't heard this criticism before
Perhaps your point is that the format is inherently stifling and lacks room for interesting discussion or exploration of novel approaches that are cleverer than the formulaic leet-code-style solution the interviewers like myself are waiting for you to write?
January 1, 2026 at 2:16 AM
Perhaps your point is that the format is inherently stifling and lacks room for interesting discussion or exploration of novel approaches that are cleverer than the formulaic leet-code-style solution the interviewers like myself are waiting for you to write?
Could you clarify what you mean by this? I've conducted quite a few interviews, the only times when I ended up writing negative feedback for someone I thought was probably pretty smart was when they failed to follow the spec and answer the question asked and then ran out of time.
January 1, 2026 at 2:11 AM
Could you clarify what you mean by this? I've conducted quite a few interviews, the only times when I ended up writing negative feedback for someone I thought was probably pretty smart was when they failed to follow the spec and answer the question asked and then ran out of time.
Also very curious where this comes from. I observed similar ideas echoed when that Malwarebytes article[1] fear-mongering about Gmail's smart features was circulating. I don't think it's new
(1) www.malwarebytes.com/blog/news/20...
(1) www.malwarebytes.com/blog/news/20...
December 22, 2025 at 2:54 PM
Also very curious where this comes from. I observed similar ideas echoed when that Malwarebytes article[1] fear-mongering about Gmail's smart features was circulating. I don't think it's new
(1) www.malwarebytes.com/blog/news/20...
(1) www.malwarebytes.com/blog/news/20...
"Broadly speaking ML generally refers to the automated identification of patterns in data or the ability of a system to learn from data without being explicitly programmed" ocw.mit.edu/courses/18-6....
December 22, 2025 at 2:06 PM
"Broadly speaking ML generally refers to the automated identification of patterns in data or the ability of a system to learn from data without being explicitly programmed" ocw.mit.edu/courses/18-6....
I'm confused as to the distinction. Could you please explain?
en.wikipedia.org/wiki/Large_l... says "The largest and most capable LLMs are generative pre-trained transformers"
en.wikipedia.org/wiki/Transfo... says "In deep learning, the transformer is an artificial neural network architecture"
en.wikipedia.org/wiki/Large_l... says "The largest and most capable LLMs are generative pre-trained transformers"
en.wikipedia.org/wiki/Transfo... says "In deep learning, the transformer is an artificial neural network architecture"
Transformer (deep learning) - Wikipedia
en.wikipedia.org
December 22, 2025 at 1:34 AM
I'm confused as to the distinction. Could you please explain?
en.wikipedia.org/wiki/Large_l... says "The largest and most capable LLMs are generative pre-trained transformers"
en.wikipedia.org/wiki/Transfo... says "In deep learning, the transformer is an artificial neural network architecture"
en.wikipedia.org/wiki/Large_l... says "The largest and most capable LLMs are generative pre-trained transformers"
en.wikipedia.org/wiki/Transfo... says "In deep learning, the transformer is an artificial neural network architecture"
Thanks for the discussion! I like Radiohead, and I don't really like social media that much so haven't put much effort into a profile.
December 19, 2025 at 6:23 AM
Thanks for the discussion! I like Radiohead, and I don't really like social media that much so haven't put much effort into a profile.
To expand the last bit, while 'distance from AGI/ASI' is unknown, partly because the measure is not well defined, we do know rather a lot about the capabilities and limits of current models - models are benchmarked independently, and there's tons of empirical usage data out in the wild.
December 19, 2025 at 6:14 AM
To expand the last bit, while 'distance from AGI/ASI' is unknown, partly because the measure is not well defined, we do know rather a lot about the capabilities and limits of current models - models are benchmarked independently, and there's tons of empirical usage data out in the wild.
Thanks, I hadn't seen that ifanyonebuildsit.com/1/will-ai-cr... - seems relevant, but TBH, I find it quite grating. They make a strong claim regarding 'fast takeoff' ("Probably") - their argument is an analogy to nuclear fission and IMHO misinformation about lack of knowledge of model capabilities.
Will AI cross critical thresholds and take off? | If Anyone Builds It, Everyone Dies
Resources and Q&A for the book If Anyone Builds It, Everyone Dies.
ifanyonebuildsit.com
December 19, 2025 at 6:08 AM
Thanks, I hadn't seen that ifanyonebuildsit.com/1/will-ai-cr... - seems relevant, but TBH, I find it quite grating. They make a strong claim regarding 'fast takeoff' ("Probably") - their argument is an analogy to nuclear fission and IMHO misinformation about lack of knowledge of model capabilities.
Anyway, thanks again for explaining what you found compelling, I did find that genuinely useful. For what its worth, I found Brian Christian's "The Alignment Problem" to be more informative and helpful as I was trying to put together my own views on the societal impact of AI/ML systems
December 19, 2025 at 5:25 AM
Anyway, thanks again for explaining what you found compelling, I did find that genuinely useful. For what its worth, I found Brian Christian's "The Alignment Problem" to be more informative and helpful as I was trying to put together my own views on the societal impact of AI/ML systems
but I guess overall, I think it flips back to Yudkowsky and Soares are making all sorts of assumptions about how AI systems *might* develop or behave, and offering no evidence for that being the primary or even a tangible risk in the real world
December 19, 2025 at 5:23 AM
but I guess overall, I think it flips back to Yudkowsky and Soares are making all sorts of assumptions about how AI systems *might* develop or behave, and offering no evidence for that being the primary or even a tangible risk in the real world
(another aside, I found the 'grown, not crafted' distinction kind of meaningless in the book as well. Agent harnesses that underlie AI systems *are* crafted. An LLM's weights doesn't inherently have access to the internet - an agent harness has to give it that ability)
December 19, 2025 at 5:21 AM
(another aside, I found the 'grown, not crafted' distinction kind of meaningless in the book as well. Agent harnesses that underlie AI systems *are* crafted. An LLM's weights doesn't inherently have access to the internet - an agent harness has to give it that ability)
I can provide lots of reasons why this isn't that significant a concern (we have lots of empirical data on the limits how frontier AI systems behave, because lots of people are evaluating them constantly! We know enough about how they work to know certain limits on their capabilities).
December 19, 2025 at 5:19 AM
I can provide lots of reasons why this isn't that significant a concern (we have lots of empirical data on the limits how frontier AI systems behave, because lots of people are evaluating them constantly! We know enough about how they work to know certain limits on their capabilities).
Aside: Yudkowsky or Soares are AI researchers or engineers by vocation. Soares might have a CS undergraduate, but neither have any real contributions in the field of AI R&D. As I recall, they provide no evidence of their claim that an AI system would develop in this way or why that might be
December 19, 2025 at 5:17 AM
Aside: Yudkowsky or Soares are AI researchers or engineers by vocation. Soares might have a CS undergraduate, but neither have any real contributions in the field of AI R&D. As I recall, they provide no evidence of their claim that an AI system would develop in this way or why that might be
I don't think we've really seen that kind of improvement or even any capacity for it. All the rapid improvements in AI systems over the last few years have still followed scaling dynamics (spend X billion on compute capacity, gathering datasets, RLHF => get X% better on various benchmarks)
December 19, 2025 at 4:58 AM
I don't think we've really seen that kind of improvement or even any capacity for it. All the rapid improvements in AI systems over the last few years have still followed scaling dynamics (spend X billion on compute capacity, gathering datasets, RLHF => get X% better on various benchmarks)
Thanks for explaining! I found a couple of things quite unconvincing about that argument. For example, it requires us to assume that AI capabilities will progress rapidly and sharply in multiple domains to the point of ungovernability before we have any ability to intervene.
December 19, 2025 at 4:54 AM
Thanks for explaining! I found a couple of things quite unconvincing about that argument. For example, it requires us to assume that AI capabilities will progress rapidly and sharply in multiple domains to the point of ungovernability before we have any ability to intervene.
Oh, may I ask which arguments you found compelling please? I ask because I personally find Yudkowsky's writing quite unconvincing, but I am familiar with this book, and I'm curious what resonated with you.
December 19, 2025 at 4:17 AM
Oh, may I ask which arguments you found compelling please? I ask because I personally find Yudkowsky's writing quite unconvincing, but I am familiar with this book, and I'm curious what resonated with you.