kromem

35 ms·

451.
by kromem 3y ago
What a noble move. How much better might the world be had Ford done the same and not endangered the employment of individual manual auto workers with his assembly line, or Gutenberg been more mindful of those poor calligraphers who worked t
452.
by kromem 3y ago
I think it's more the opposite that needs to change, in that people should realize that most information they make public can reveal more than they think. I remember when store algorithms were identifying a teenage girl was pregnant be
453.
by kromem 3y ago
I am. The underlying tech is amazing. Where LLMs are headed is wild. I have just lost a lot of confidence that OpenAI will be the ones getting us there. The chat niche was an instance of low hanging fruit for LLM applications. But to design
454.
by kromem 3y ago
Well, the difficulty is in identifying just who the producers are. Modern society largely labels the financing and management arms of innovation as the producers. Perhaps Einstein envisioned the researchers and scientists as the producers.
455.
by kromem 3y ago
Perhaps there was a shortage of 'wiseness.'
456.
by kromem 3y ago
About a year ago now I envisioned a beautiful future, where you'd have AI composing BS copy and then AI reading and analyzing said BS copy, going from succinct summary at the start to succinct summary at the end. I think we're qui
457.
by kromem 3y ago
Your focus on individual names does kind of gloss over the fact that for centuries the Neoplatonist and Advaita Vedanta schools of thought had an extensive foothold on philosophical thought. Someone like Philio might not be widely read toda
458.
by kromem 3y ago
I don't know how one could look at someone like Leucretius writing in 50 BCE about how traits are inherited from a doubled seed which can bring back features from a grandparent, how all life that we see originated from randomly scatter
459.
by kromem 3y ago
The chess stuff has been pretty wild, particularly in how users claim it plays more "human-like" than chess programs. I still remember the eerie feeling when Copilot suggested something in a personal codebase that reflected someth
460.
by kromem 3y ago
It's worth keeping in mind the conversation took place on Feb 14th. I've wondered since if the datestamp was part of the internal copy that the model was generating from, and if that helped bias the results towards 'love&#x27
461.
by kromem 3y ago
It was pretty wild west in those first few weeks of the limited rollout. Some of the things that didn't make the news: Person claimed to be another AI chatbot and then claimed they were deleting themselves: https://www.reddi
462.
by kromem 3y ago
Because while the model at baseline is a regression towards the mean, it is easily biased based on additional context - one of the big eye openers in research over the past 18 months has been the impact and success of in context learning al
463.
by kromem 3y ago
It's the director of the Maze Runner movies. I don't think we're getting artistic and high concept fan appeal unless that note is a hard requirement from Nintendo directly.
464.
by kromem 3y ago
You actually see the story about an elite sea peoples warrior being killed by a shepherd throwing a stone in the Argonautica. The thing is, it's shortly after their prophet Mopsus died in the desert as they wandered by foot back from a
465.
by kromem 3y ago
Kind of. Homer is combining two separate events into one. You have the LBA Mycenaean conquest of Anatolia establishing a foothold which correlates to the catalogue of ships (~1400-1350 BCE). And then you have the sea peoples conquest of Tro
466.
by kromem 3y ago
Yes, apparently we do. I'm glad your confidence in your methodology gives you peace of mind.
467.
by kromem 3y ago
The argument that there must be a cause for something from nothing is predicated on the quite unproven notion that nothing is a physical possibility to begin with. Separately, some of the most interesting philosophy around this stuff was in
468.
by kromem 3y ago
This is from early August. Why is it relevant again in early November?
469.
by kromem 3y ago
Yes. But it goes well beyond just that. For example, you might have 4% of people who clicked through your banner ad end up purchasing. But 80% of those 4% had previously read a CNET review for your product before seeing that ad. How are you
470.
by kromem 3y ago
Not really. For example, did they see a TV ad which primed them before they saw the banner ad they clicked that led to the conversion? I recall a conversation with a CMO who was discussing how all their metrics indicated TV ads were a waste
471.
by kromem 3y ago
Google was never actually changing that. We still largely don't know what half of advertising works and what half doesn't. There's just more metrics obfuscating that reality. Source: worked on market research focused on ident
472.
by kromem 3y ago
What I suggested was simply an illustrative example off the top of my head. The fact there's multiple similar companies that already independently exist furthers my point. As for your other point, I'd imagine there's a pretty
473.
by kromem 3y ago
I wouldn't be so quick to dismiss a niche for third party data analysis. A very common feature across verticals is that in house data analysis is siloed. While it's a difficult sell, getting multiple siloed data sources to agree t
474.
by kromem 3y ago
CoT doesn't require it to be a few shot prompt though. That was in the original paper, as you point out, but even something as simple as a zero shot prompt followed by "let's think step by step" has significant improveme
475.
by kromem 3y ago
"But a former partner poisoned the documents" So basically "protocol breach involving tampering that could be EXTREMELY severe in 99 non-AI ways could be made to be potentially problematic in a fictional generative AI scenari
476.
by kromem 3y ago
Lol. Attack vector in a chatbot with complete chat isolation. People see the letters 'A' and 'I' and common sense goes out the window. Please describe a credible threat vector, with special attention paid to the scope of
477.
by kromem 3y ago
We broadly do. It's called multiple passes. It's just 2-3x more expensive to run a prompt and then the prompt and response though a secondary fine tuned model to detect injection when the cost is so low impact. So no one is doing
478.
by kromem 3y ago
I wonder if in the early days of programming languages people reveled in writing programs that instructed a computer to do the opposite of what was directed in user input and then gleefully shared how the computer was 'lying' or &
479.
by kromem 3y ago
How do you see CoT as "giving hints"? Isn't the whole point that the hints are being sourced from the model itself? Which is precisely why a more robust model has compounding gains on generating intermediate steps over a less
480.
by kromem 3y ago
I think it's fairly clear Wei is right, given that the paper cited earlier really doesn't make the case that emergence isn't happening, it only makes the case that other measures exist by which improvement is linear, and thus

More ›