MacsHeadroom

11 ms·

91.
by MacsHeadroom 3y ago
You can get Bilberry spread and syrup at Ikea in the U.S. It's very good. But it probably also has added sugar.
92.
by MacsHeadroom 3y ago
Perplexity says ~66% of the internet is hosted on U.S. infrastructure (AWS, Azure, Google Cloud, etc.).
93.
by MacsHeadroom 3y ago
Four V100s will do. They're about $1k each on ebay.
94.
by MacsHeadroom 3y ago
80GB in 4bit. But because it only activates one expert at a time, it can run on a fast CPU in reasonable time. So 96GB of DDR4 will do. 96GB of DDR5 is better.
95.
by MacsHeadroom 3y ago
It's open source, you can just download and run it for free on your own hardware.
96.
by MacsHeadroom 3y ago
At https://poe.com/
97.
by MacsHeadroom 3y ago
There is no way to anonymize transcripts where I and others refer to me by name and reference uniquely identifiable information. What happens when the AI provider is served with a warrant or subpoena for transcripts mentioning MacsHeadroom
98.
by MacsHeadroom 3y ago
And Claude Sonnet is great. It's what I recommend people use if they don't want to pay for an LLM or run one locally. But Claude Sonnet is not open source and can't be run on your own hardware like Wizard.
99.
by MacsHeadroom 3y ago
It means "home/house spirit" and from context sounds like they're referring to Google Home or Alexa.
100.
by MacsHeadroom 3y ago
That's 91% cheaper than a human therapist (~$200/hr). That's 52 weekly one hour "therapy" sessions for the price of one month of weekly sessions with a human.
101.
by MacsHeadroom 3y ago
I love art made by non-human intelligences. I especially love how it can transcend and redefine loved mediums by combining them in surreal ways that are otherwise quite difficult to obtain. Algorithmic exploration of mediums outpaces mere m
102.
by MacsHeadroom 3y ago
> the copyright system creates a (relatively) shared understanding between creators and consumers that, by default, content cannot be redistributed, remixed, adapted or built upon without creators' consent. This shared understanding
103.
by MacsHeadroom 3y ago
Subjective experience is the awareness of an internal representation.
104.
My binary vector search is better than your FP32 vectors(blog.pgvecto.rs)
3 points by MacsHeadroom 3y ago|0 comments
105.
by MacsHeadroom 3y ago
Were these private firms funded by the government?
106.
by MacsHeadroom 3y ago
Even pocket computers (smartphones) are already better than billion dollar supercomputers from decades past. What is your point?
107.
by MacsHeadroom 3y ago
> bring down the average laborers value. Labor has never been to afford so much luxury as the modern day.
108.
by MacsHeadroom 3y ago
Q & A are not part of pretraining.
109.
by MacsHeadroom 3y ago
All court orders are warrants. That is the most basic definition of a warrant. So a subpoena issued by the court is warranted. A subpoena not issued by a court but accompanied by a separate warrant for the same information constitutes a war
110.
by MacsHeadroom 3y ago
Yes, but if presented with a warrant for your property (ex. subscriber information) you must comply or be charged with obstructing justice or impeding an investigation. When presented with an administrative subpoena not backed by a warrant
111.
by MacsHeadroom 3y ago
This is not how ML works generally nor LLMs like ChatGPT specifically. What you're describing sounds like RLHF, which changes the style of responses and impacts things like refusals but does not add to a model's intelligence (in f
112.
by MacsHeadroom 3y ago
I think the difference between 2000 and 2020 is already greater than the difference between 1900 and 2000. Smartphones, video conferencing, remote work, social media, are pervasive and consume the majority of most peoples waking lives.
113.
by MacsHeadroom 3y ago
This is exactly right. Model collapse does not exist in practice. In fact, LLMs trained on newer web scrapes have increased capabilities thanks to the generated output in their training data. For example, "base" pretrained models
114.
by MacsHeadroom 3y ago
Energy usage is good, actually. Energy scarcity and dirty energy are bad. But energy usage is good. We should strive to use and produce orders of magnitude more (clean) energy.
115.
by MacsHeadroom 3y ago
Model collapse is basically a myth and is a joke in the ML community. The assumptions for the model collapse paper do not hold in the real world even when training on uncrurated generated data. In fact, LLMs of equal size trained on newer w
116.
by MacsHeadroom 3y ago
Use a debrid provider or seedbox to download the torrent. They torrent it for you and then you direct download from them. Should cost $10 or less.
117.
by MacsHeadroom 3y ago
Fair use is an exception to infringement. Use which is fair is non-infringing. For example, Google books containing a searchable copy of every book ever written is fair use, as is Google's cache containing every news article and web pa
118.
by MacsHeadroom 3y ago
This paper is a joke in the ML community. Model collapse is not real. This paper made some seriously faulty assumptions in their contrived lab experiment. State of the art models are trained entirely or largely on "synthetic" gene
119.
by MacsHeadroom 3y ago
GPT-4 with code interpreter is better at math than elite Math undergrads.
120.
by MacsHeadroom 3y ago
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...

More ›