mlb_hn

5 ms·

1.
by mlb_hn 1y ago
Our GCP is down
2.
by mlb_hn 2y ago
nice overview of progress over time. are there quant metrics for the sim capabilities or is it mostly vibes?
3.
by mlb_hn 2y ago
he was good people
4.
Accelerating Development with LLMs by Hallucinating Functions(engineering.hyperwriteai.com)
3 points by mlb_hn 4y ago|0 comments
5.
by mlb_hn 5y ago
yeah, that's a very big caveat - haven't checked neo 20b yet. I've had a hard time getting the AI21 models to use it and those are also pretty big so it's interesting why sometimes it works and sometimes it doesn't.
6.
by mlb_hn 5y ago
definitely. also works on text translation/comprehension like emojis! https://aidungeon.medium.com/introducing-ai-dungeon-translat... . For actual benchmarks, scratchpad improves GPT-Davinci WIC from 50% accuracy (chanc
7.
by mlb_hn 5y ago
I get the tokenization argument and it may influence it a bit, but I suspect the n-digit math issue has to do more with search the way it samples (in the bpe link gwern references some experiements I'd done with improving n-digit math
8.
by mlb_hn 5y ago
Couple things there where you can see if it improves with the prompt/formatting. E.g. with Davinci (and J a bit but didn't test too much) you can get bette results by: - Using few-shot examples of similar length to the targets
9.
by mlb_hn 5y ago
An alternative of that is as Oppenheimer once said, sometimes things are secret because a man doesn't like to know what he's up to if he can avoid it
10.
How We Accidentally Gave Our Bots Their Personalities(aidungeon.medium.com)
2 points by mlb_hn 6y ago|0 comments
11.
by mlb_hn 6y ago
This has been a systemic issue reported on for years; e.g. reported by the Intercept in 2017 [1] and Atlantic in 2019 [2]. Not really made clear from the story considering the Economist headline is almost identical to the Atlantic one. [1]
12.
by mlb_hn 6y ago
I think that goes back to a Karpathy quote [1], don't know where he got it from [1] https://twitter.com/karpathy/status/961307010246492160
13.
by mlb_hn 6y ago
My thinking there wasn't because of BPEs, I think it's a graph traversal issue.
14.
by mlb_hn 6y ago
Also, better prompt design if you have make implicit meaning explicit can improve the WiC score ( http://gptprompts.wikidot.com/linguistics:word-in-context ) and ANLI score ( http://gptprompts.wikidot.com/lingu
15.
by mlb_hn 6y ago
Just priming an immediate availability response is likely going to get poor results. On the other hand, this does bring up an important point, which is that few people have been systemically trying to figure out how to get it to reason thro
16.
by mlb_hn 6y ago
Great article. It's worth considering the Japanese reaction to the bomb - people thought it was cluster munitions or other conventional explosives leading to them taking sub-optimal follow-on steps. While the physicists figured it out,
17.
by mlb_hn 6y ago
I don't think it's going to be about a single prompt; reverse engineering multiple prompts interacting with themselves is hard. There's a lot of cool things to be done with: (a) creating a pipeline of prompts that combine out
18.
by mlb_hn 6y ago
I think it would be jumping to conclusions to say because the military has vulnerabilities that the entire military industrial complex is a fraud. Pieces of it certainly are sub-optimal and arguably fraudulent though such as DCGS (the &quot
19.
by mlb_hn 6y ago
That's the dataset used in the paper.
20.
by mlb_hn 7y ago
Leadership in the US Government wasted a lot of time trying to play down the pandemic [1] as did large (conservative) news organizations [2]. Both are now trying to blame China, arguably to shift blame from themselves. It didn't need t
21.
by mlb_hn 7y ago
I have a WSJ subscription. I don't know of any easy way around their paywall, makes it one of the special cases for scraping because you would actually need to login.
22.
by mlb_hn 7y ago
I'm not sure I would agree that the article doesn't matter that much. It's an argument that while the FCC's going along with what the courts decided, they're not necessarily doing so in good faith because they tried
23.
by mlb_hn 7y ago
What do you mean by dictate what you see? Aside from the sponsored content/ads, isn't it just a prioritized queue of what your Facebook friends posted?
24.
by mlb_hn 7y ago
Is he wrong? Like a telco, Facebook allows people to communicate with one another without going through another intermediary. However, it's not one-to-one like your phone, it's one-to-many. Like a newspaper the messages are broadc
25.
by mlb_hn 7y ago
Here's Facebook's business help link for how to upload and use point of sale and other offline data: https://www.facebook.com/business/help/1142103235885551?id=5...
26.
Social Media Companies Are Failing to Combat Inauthentic Behaviour Online(stratcomcoe.org)
1 points by mlb_hn 7y ago|0 comments
27.
by mlb_hn 7y ago
Masquerading as Red Cross can be a grave breach of Article 37 of the 1977 Additional Protocol I [0] [0] https://ihl-databases.icrc.org/ihl/WebART/470-750111 (paragraph 3.f)
28.
by mlb_hn 7y ago
Is there a legal basis for that argument or is that your opinion?
29.
by mlb_hn 7y ago
Espionage/sabotage is not a war crime https://ihl-databases.icrc.org/customary-ihl/eng/docs/v2_rul...
30.
by mlb_hn 7y ago
The money's one thing, but my concern is that the same ad-fraud tech used to fake all those views can fake views to actual news sites (and influence coverage via analytics). If it can beat Google and Amazon's fraud detection, how

More ›