cosentiyes

They're the only non-nvidia accelerator used to train state of the art large language models at scale?

> matlab level of usefulness

that's a little harsh :D

I have the same problem on iMessage. iCloud reports 7+gb of message storage being consumed, but each of my devices and my top conversations only report 1-2gb of content. I haven't found a way to force sync the missing content so I'm stuck paying the $1/month plan.

The actual paper title is "Mitochondrial DNA damage triggers spread of Parkinson’s disease-like pathology": https://www.nature.com/articles/s41380-023-02251-4. Parkinson's is a notoriously complex disease with multiple etiologies and subtypes. While this is particularly exciting to see (I have a loved one with PD), the EurekAlert article really misses the critical "PD-like" modifier.

I was under the impression that relay's phone number masking is voip (similarly to google voice). In my experience, many signups do not allow voip numbers.

for anonymity, yes (if using a "private" snowflake). for daily use or basic firewall circumvention, not really. as others have mentioned, it is more convenient and faster to just use shadowsocks to a digital ocean droplet or other vm.

> Solve the barriers/limitations

I don't think this is possible when you can't pool memory on the 40 series retail GPUs.

Agreed, having a bunch of lifetime customers without a source of recurring revenue also incentivizes telemetry, selling user data, etc. Jellyfin is also a strong enough competitor that many lifetime accounts would switch if new fees appear.

I'm afraid that unraid will run into similar problems eventually.

Yes, the EU is doing what needs to be done (criminalizing end-to-end encryption) /s

"The Sacred and The Profane: The Nature of Religion" by Mircea Eliade and "Eight Theories of Religion" by Daniel Pals aren't focused on a single religion but provide nice overviews on its history and evolution.

> You can make every promise in the world, because nobody can verify them. You don't even have to know what you're doing, because again, nobody can verify what you say. It is 100% snake-oil.

I wonder how the author feels about audits? At the end of the day, even if mullvad is logging, I trust them with my data far more than xfinity/comcast.

[1] https://mullvad.net/en/blog/2021/1/20/no-pii-or-privacy-leak...

it's also on youtube: https://www.youtube.com/watch?v=L_Guz73e6fw

"We can’t listen to your queries and no other party can. Privacy is built by design in our solution."

I don't really understand the technical solution behind this statement. Isn't this just e2e encryption up until you decrypt the query to pass to the LLM? ML operating via homomorphic encryption is very far away and OpenChatKit is just a standard self-hosted LLM. This seems more like "self host on semi trusted azure compute that isn't owned by openai and the model performance will be far worse than gpt4"?

I'm under the impression that the vast majority of people use their cloudtop over ssh and a remote vscode connection. I only ever use a remote gui for a few minutes every X weeks to do chrome-based auth for an nonstandard workflow.

They're literally flagging one "sanctioned" way to provide signal as an example in a resume advice post--how is that a waste of time? This lets people infer what other types of "optional" content might provide signal: publications, blog posts, etc. Transparency in the application process is fantastic and it's wild to me that people are upset about a company posting some ways that applicants can make themselves stand out. This is order of magnitudes better than VC, finance, etc.

Slightly disagree, they're just providing one potential way to signal experience. I'm sure there are other ways not listed (firth author publications from tier 1 conferences or journals) that are more valuable. I think you're right that this could be used as a tie-breaker, all other things being equal.

To be fair, they explicitly say that github/so/etc is optional:

We understand that not everyone has time to contribute to public discourse, so #2 is optional.

they could if everyone donates!

lichess.org >>> chess.com. I encourage everyone who can afford to donate to do so!

I haven't seen anything on chess.com's stack before. For gp, previous hn discussion on lichess and scala 3: https://news.ycombinator.com/item?id=33865932

scale.ai lays off 20% across roles

this is new (as of may 2022) and doesn't always assume the correct epub encoding, which can insert artifacts (eg see https://github.com/janeczku/calibre-web/issues/2508).

Internal combustion engine

Yes, this is a companion to https://arxiv.org/abs/2209.14958: "We share many reflections in our published pre-print (Mirowski, Mathewson et al (2022) “Co-Writing Screenplays and Theatre Scripts with Language Models: An Evaluation by Industry Professionals.”)"

Explicitly flagging this llm as a tool for (potentially) writing fiction is a nice way to work on domain-specific text generation while avoiding many of the critiques of Galactica.

A nice first step, but somewhat frustrating that they don't include pre-2020 models. A fan in my 2018 mbp has recently started to act up and "reputable" sellers (ifixit) are currently out of stock.

There is also a research group: https://health.google/health-research/genomics/

Wrong, #1 in all categories.

they're not testing for memorization skills but for "desire to join the community"

You were.