Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
SrslyJosh
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
9 ms
·
91.
▲
by
SrslyJosh
2y ago
Well, challenge number one is having a consumer protection agency that's legally empowered to pull products that are defective by design from the market. Second challenge is precisely defining what is and is not allowed, but given the
92.
▲
by
SrslyJosh
2y ago
This would be more interesting if there was actually a proposed solution.
93.
▲
by
SrslyJosh
2y ago
Please don't.
94.
▲
by
SrslyJosh
2y ago
AI bros: "Why does it matter whether [a program that people want to use to teach children] can [tell whether one number is bigger than another]?" =)
95.
▲
by
SrslyJosh
2y ago
None of the videos played for me in Safari 16. =(
96.
▲
by
SrslyJosh
2y ago
> Yeah, I agree. Being easy to use isn't the indented use-case for gRPC. Sick burn. I like it, especially since most use of gRPC seems to be cargo-culting.
97.
▲
by
SrslyJosh
2y ago
>> Also, it doesn’t pass my “send a friend a cURL example” test for any web API. > Well yeah. It’s not really intended for that use-case? Until $WORKPLACE is invaded by Xooglers who want to gRPC all the things, regardless of whethe
98.
▲
by
SrslyJosh
2y ago
Relevant link: https://www.wired.com/story/microsoft-windows-recall-privile... TL;DR: Recall's DB can be accessed by any malicious app running with user-level privileges. =)
99.
▲
by
SrslyJosh
2y ago
And then you do search to check that the output reflects reality, right? If not, good luck.
100.
▲
by
SrslyJosh
2y ago
Fatalities aren't the whole story. > The 2023 records, newly disclosed by the Occupational Safety and Health Administration, also show that injury rates at some SpaceX facilities grew worse than those the company had reported in 202
101.
▲
by
SrslyJosh
2y ago
Israel: The "ally" that acts like an enemy. =) (Alternately, the client state that controls its patron.)
102.
▲
by
SrslyJosh
2y ago
What does this have to do with the US' unconditional support of an apartheid regime in the middle east?
103.
▲
by
SrslyJosh
2y ago
> I rarely trust code it generates (at least for the language I mainly work in) as i’ve seen it make some serious mistakes (eg: using keywords in the language that don’t exist) It's only a mistake from your perspective. The model ju
104.
▲
by
SrslyJosh
2y ago
> are AI safety folks building models of this pattern? First you need to ask if AI "safety folks" actually understand the technology, and if they are thinking about it objectively. If they believe that we're a few years aw
105.
▲
by
SrslyJosh
2y ago
> LLMs still hallucinate Keep in mind that there's no difference between what happens inside a model when it "hallucinates" vs. when it generates "correct" output. It's the exact same process.
106.
▲
by
SrslyJosh
2y ago
> hallucinated the answer An LLM that's "hallucinating" is doing the exact same thing as an LLM that's generating "correct" output. =)
107.
▲
by
SrslyJosh
2y ago
> Policy tuning is one way to remove it, but there's some research on applying policies like "this text is a lie" during pretraining instead of trusting all of it equally. Who's gonna vet all the training material?
108.
▲
by
SrslyJosh
2y ago
I'm not sure myself. I didn't want this to be the second comment on the submission so I'll say it now: I got TimeCube vibes from this.
109.
▲
by
SrslyJosh
2y ago
Given: 1. The plot of "Her" (guy falls in love with synthesized voice, played by Johansson) 2. Altman's affinity for the film (the article says he's called it his "favorite movie") Reaching out to Johansson abo
110.
▲
by
SrslyJosh
2y ago
I don't think this is an apples-to-apples comparison. The movie producers didn't produce a simulation of Britney's voice and attempt to sell access to it. However you feel about an probably-unapproved celebrity cameo in a mov
111.
▲
by
SrslyJosh
2y ago
The models look awful. If only there was a place where you could go to download human-created 3d models, either for free or a nominal fee. /s
112.
▲
by
SrslyJosh
2y ago
The term "hallucinations" incorrectly anthropomorphizes LLMs and falsely assumes that there is a difference between "correct" and "incorrect" output. It's all the same to the model. LLMs are probalistic an
113.
▲
by
SrslyJosh
2y ago
You think he's a fool, but remember that extra space in the cybertruck battery pack? That's where the CyberCoal™ add-on pack will go, which will allow customers to expel huge clouds of choking black smoke at the push of a button.
114.
▲
by
SrslyJosh
2y ago
> There is Tesla AI for example that is building Robots etc. Musk's competitor to OpenAI (x.AI ?) has raised $6 billion at a $18 billion valuation. This is just giving Musk more runway to make promises that Tesla will fail to delive
115.
▲
by
SrslyJosh
2y ago
America is too expensive. The entire economy has been shifting away from creating new useful things and towards what I think of as extraction. A couple examples: - Car companies paywalling hardware and software features while selling data (
116.
▲
by
SrslyJosh
2y ago
Not sure that's a major issue for current customers; I saw a guy careening through my office parking lot at road speed just yesterday. No hands on wheel, staring at his phone.
117.
▲
by
SrslyJosh
2y ago
Congrats to OpenAI (and the rest of the LLM bros) for creating negative incentives for sharing knowledge.
118.
▲
by
SrslyJosh
2y ago
> Generally any white collar criminal does not need to go to "big boy jail" Why not? Are they special because they hurt people indirectly instead of directly? > a hypothetical executive who was convicted of ordering a hit We
119.
▲
by
SrslyJosh
2y ago
That's a lot of anthropomorphizing, but I don't see anything to back up your claims.
120.
▲
by
SrslyJosh
2y ago
> I'm a high school student with a passion for tackling misinformation online. Inspired by the need for more reliable content verification tools, I decided to create Factful. It's an AI-powered web app I stopped reading here be
More ›