3 ms·

You notice how there's no patents being thrown about, in all this recent hype? NSA Echelon program covered the space in the 90s. Note their ongoing silence. I

You notice how there's no patents being thrown about, in all this recent hype? NSA Echelon program covered the space in the 90s. Note their ongoing silence. I strongly suspect that LLMs will not be the miracle (or disaster) that they're currently being sold as.

PaulHoule 4y ago
Edward Snowden's bosses boss gave a rather good seminar on text analysis systems at a Semantic Technology Conference (that I attended) at the Union Square Hilton in San Francisco about a week before the Snowden "revelations" broke. There was a time I was doing all sorts of sales and networking calls and found there were a lot of people in the Research Triangle Park area who had forgotten more about text analysis than anyone in the Bay Area knew. The "secret" was not one great idea such as transformer models but rather the social and technical organization to break up difficult problems into easy and difficult parts and farm those out to very large groups of people. A person getting started in text analysis might look at Apache UIMA and ask "Where's the beef?" because UIMA doesn't really do anything all but rather is a framework for getting 10,000 people pulling in the same direction. IBM Watson did have the "big idea" of using probability scores to combine multiple question answering strategies but the real "secret" again was that they had the equivalent of an army batallion work on it. If you don't believe me, try driving from the BWI airport to Washington DC at rush hour and witness for yourself America's worst traffic jam when the day shift at Ford Mead (the NSA headquarters) gets off work. If there's one "secret" about secrets it that it gets increasingly hard to keep a secret the more people that know about it. (And I'll say yes, I know those details make me sound paranoid, I am just as paranoid as I sound, because if I told you I wasn't paranoid that would just convince you all the more that I am paranoid.)
- h2odragon 4y ago
  The question is, are you paranoid enough?
JohnFen 4y ago
> I strongly suspect that LLMs will not be the miracle (or disaster) that they're currently being sold as. I am 99.9% sure that this is the case. The hype level is insane right now, and people (both pro and anti) are commonly asserting patently absurd things as fact. In the end, I predict, we'll find that this is a powerful and useful tool, but it won't be earth-shaking.
ogogmad 4y ago
> NSA Echelon program covered the space in the 90s How? > I strongly suspect that LLMs will not be the miracle (or disaster) that they're currently being sold as. Why do you think this? See Heuristics That Almost Always Work: https://astralcodexten.substack.com/p/heuristics-that-almost-always-work https://astralcodexten.substack.com/p/heuristics-that-almost...

PaulHoule 4y ago

Edward Snowden's bosses boss gave a rather good seminar on text analysis systems at a Semantic Technology Conference (that I attended) at the Union Square Hilton in San Francisco about a week before the Snowden "revelations" broke. There was a time I was doing all sorts of sales and networking calls and found there were a lot of people in the Research Triangle Park area who had forgotten more about text analysis than anyone in the Bay Area knew. The "secret" was not one great idea such as transformer models but rather the social and technical organization to break up difficult problems into easy and difficult parts and farm those out to very large groups of people. A person getting started in text analysis might look at Apache UIMA and ask "Where's the beef?" because UIMA doesn't really do anything all but rather is a framework for getting 10,000 people pulling in the same direction. IBM Watson did have the "big idea" of using probability scores to combine multiple question answering strategies but the real "secret" again was that they had the equivalent of an army batallion work on it. If you don't believe me, try driving from the BWI airport to Washington DC at rush hour and witness for yourself America's worst traffic jam when the day shift at Ford Mead (the NSA headquarters) gets off work. If there's one "secret" about secrets it that it gets increasingly hard to keep a secret the more people that know about it. (And I'll say yes, I know those details make me sound paranoid, I am just as paranoid as I sound, because if I told you I wasn't paranoid that would just convince you all the more that I am paranoid.)

h2odragon 4y ago
The question is, are you paranoid enough?

JohnFen 4y ago

> I strongly suspect that LLMs will not be the miracle (or disaster) that they're currently being sold as. I am 99.9% sure that this is the case. The hype level is insane right now, and people (both pro and anti) are commonly asserting patently absurd things as fact. In the end, I predict, we'll find that this is a powerful and useful tool, but it won't be earth-shaking.

ogogmad 4y ago

> NSA Echelon program covered the space in the 90s How? > I strongly suspect that LLMs will not be the miracle (or disaster) that they're currently being sold as. Why do you think this? See Heuristics That Almost Always Work: https://astralcodexten.substack.com/p/heuristics-that-almost-always-work https://astralcodexten.substack.com/p/heuristics-that-almost...