gardnr

6 ms·

91.
by gardnr 11mo ago
What does it do that Opus doesn't do?
92.
by gardnr 11mo ago
Telegram, which ostensibly claimed to provide e2e but really only did in very specific circumstances? My right wing uncle is still bitter about that. Then there's the rolling over the founder did after getting pulled up by Interpol. I&
93.
by gardnr 11mo ago
I'm not sure if you're jabbing at the concept of American supremacy, or Hinton's idea, or my position. I don't live in the USA right now, but I am happy to participate in conversation. That's why I am here. Can you
94.
by gardnr 11mo ago
I get a lot from seeing the person talk vs reading a summary. I have gone back and watched a lot of interviews and talks with Ilya. In hindsight, it is easy to hear the future ideas in his words at the time. That said, I use AI summaries fo
95.
by gardnr 11mo ago
He says the current models generalize dramatically worse than people: https://youtu.be/aR20FWCCjAs?t=1501 Then, he starts to talk about the other ideas but his lawyers / investors prevent him from going into detail: h
96.
by gardnr 11mo ago
I thought about the huge pile of hard drives in Utah this morning. The TLAs in the USA have a metric shit ton of data that _should_ not be used but _could_ be used. Even still, we need evolutions in model architecture to get to the next lev
97.
by gardnr 11mo ago
The gains were on benchmarks. Ilya describes why this is a red herring here: https://youtu.be/aR20FWCCjAs?t=286
98.
by gardnr 11mo ago
This is an AI generated article based on real interviews. Watch the original Sutskever interview: https://www.youtube.com/watch?v=aR20FWCCjAs And LeCun: https://www.youtube.com/watch?v=4__gg83s_Do
99.
by gardnr 11mo ago
How does it compare and contrast to DNSCrypt? https://github.com/DNSCrypt
100.
Show HN: OctoDNS, Tools for managing DNS across multiple providers(octodns.readthedocs.io)
31 points by gardnr 11mo ago|3 comments
101.
by gardnr 11mo ago
Gemini 3 Deep Think gets 45.1% on ARG-AGI-2 Gemini 3 Pro gets 31.1% on ARG-AGI-2 https://arcprize.org/leaderboard
102.
by gardnr 11mo ago
A discussion on models "figuring out" things: https://www.youtube.com/watch?v=Xx4Tpsk_fnM (Forbidden Technique)
103.
by gardnr 11mo ago
GLM 4.6 isn't a "fast" model. It does well in benchmarks vs Sonnet 4.5. Cerebras makes a giant chip that runs inference at unreal speeds. I suspect they run their cloud service more as an advertising mechanism for their core
104.
Ask HN: Lawyers of HN, how do you deal with AI slop?
8 points by gardnr 11mo ago|2 comments
105.
by gardnr 11mo ago
It might be better in its own "Ask HN" post. If you want to get into AI topics: start learning about LLMs, embeddings, vector stores, datasets, fine-tuning, evals, and how to use CLI dev tools like Claude Code. Andre Ng helped bui
106.
by gardnr 1y ago
Obligatory Fireship video: https://www.youtube.com/watch?v=yl0YWA2K2B0&t=40
107.
by gardnr 1y ago
I appreciate the insight in this post and I would argue that we need the right amount of loss to enable the next level. Trying to achieve "lossless representation" will be very expensive and may not bring the desired outcome. Huma
108.
by gardnr 1y ago
Can anyone please explain this in simple terms for a relative outsider?
109.
by gardnr 1y ago
Does PyTorch support other acceleration? I thought they just support Metal.
110.
by gardnr 1y ago
The model weighs 1.5GB [1] (the q4 quant is ~500MB) The demo is impressive. It uses reference audio at inference time, and it looks like the training code is mostly available [2][3] with a reference dataset [4] as well. From the README: &gt
111.
by gardnr 1y ago
As a customer, I sent an email saying that it felt like a cash-grab instead of a genuine attempt to improve customer experience. Pretty sure that email single-handedly push the needle on their decision. Hah!
112.
by gardnr 1y ago
Has there been a more recent event, or are you referring to Alexandria?
113.
by gardnr 1y ago
In other words? > The only way that would work if Google could decrypt the message!
114.
by gardnr 1y ago
Notion seems to be working: https://www.notion-status.com/
115.
by gardnr 1y ago
TBF, they all believed that scaling reinforcement learning would achieve the next level. They had planned to "war-dial" reasoning "solutions" to generate synthetic datasets which achieved "success" on complex r
116.
by gardnr 1y ago
Most people might remember it from the headlines: > In June 2022, LaMDA gained widespread attention when Google engineer Blake Lemoine made claims that the chatbot had become sentient. The scientific community has largely rejected Lemoin
117.
by gardnr 1y ago
Then there's the billing dashboards... It's amazing how they can show useless data while completely obfuscating what matters.
118.
by gardnr 1y ago
It is more complex than just a threat in our environment. Huberman has done a few episodes on it: https://www.hubermanlab.com/episode/tools-for-managing-stres...
119.
by gardnr 1y ago
Consider: https://doi.org/10.1016/j.bbi.2016.02.024
120.
by gardnr 1y ago
The logic doesn’t follow. If you already pay for something with taxes, why would you want to pay for it again? Maybe something like a toll road so you can get there faster; or another private service that is better than the public one. By y

More ›