forgingahead

Don't they mean "illegal immigrants"? When did "migrant" come into the lexicon to distort the discussion?

I've seen some repos on Github host the models on "Releases". Not sure how viable that is long term, but definitely easier to download and access than some random Google Drive link which is typical for a lot of other repos.

The bottom line seems to be: Be careful of agencies!

That's exactly my point. In high-trust, homogenous societies, you don't necessarily need harsh laws to enforce social norms and "civilised" behaviour. In more heterogenous societies, you have low-trust so you need tough laws to enforce the societal "norms".

Japan is the former and Singapore is the latter. Both end up in good places to live, but one has to have the full discussion about the reasons and causes for each.

Japan is highly homogenous - their immigration policy would be decried by many liberals in the West as "racist". It is a massive factor in how their society is run, and why they are so high-trust that they don't need strong enforcement mechanisms (in some cases. Don't be a Western car exec there for example).

There is a show on Netflix called "Old Enough!". It's worth watching to see how some of the high-trust works in practice, and also to perhaps think through if we could be comfortable with the same in our own neighbourhoods and countries. My guess for most people in the Anglo West, the answer would be "No".

Could be said about any country that has allowed their yellow newspapers to take over the discourse (and I include most US broadsheets in this category).

What is anti-vax about it?

What a lovely written article, and a nice throwback to that time of the "earlier" internet. I also enjoyed the photo of the author in the motel hunched over his laptop; certainly rings a few bells for myself (and perhaps others here as well).

I remember arguing with my family as I was trying to play Ultima Online on our 28.8k connection and someone wanted to use the phone. Salt in the wound was when the phone call wasn't urgent but rather just a normal chin-wag with a friend. Was always a thrill to discover if during that sliver of time after school and sports, and before homework and dinner, and nobody was using the phone (and I hadn't used up my current allotment of our shared home PC as well), I could log into UO Sonoma and play for a short but glorious time, and hopefully avoid getting grief-killed as well.

Really fun times and good memories.

Zuck is a total killer. What better way to fight Google and Microsoft than to effectively spawn thousands of potential competitors to their AI businesses with this (and other) releases. There will be a mad scramble over the released weights to develop new tech, these startups will raise tons of money, and then fight the larger incumbents.

This is not charity, this is a shrewd business move.

So - Drescher 2024 then?

Seinfeld did a great bit on car rentals: https://www.youtube.com/watch?v=4T2GmGSNvaM

Need all Python ML repos to be migrated to Ruby equivalents ASAP.

Pretty incredible. We really take for granted the modern connectivity and access of goods and people compared to what human beings had to do for most of history.

Tangentially related: Clive Cussler[0] books are good fun if you're looking for a fiction series about shipwrecks and other nautical adventures.

[0]https://www.amazon.com/stores/Clive-Cussler/author/B000APJ4L...

Yes Python is incredibly annoying to use. Their dependency management is a total mess, and it's incredible how brittle packages if there are even minor point changes in versions anywhere in a stack.

This is a good insight, but the interesting counter-weight to the reality you describe is that many people who want "remote" and the independence of work, don't really want the same "benevolent dictatorship" either.

* Don't use a single embedding per content item, use multiple to increase retrieval quality

Can you share some specific examples of what you mean by this? How would you process specific info types (eg: news article, or web page, or product catalogue data) this way, and how would you handle retrieval that makes the quality "better"?

*Edit: Thanks for all replies so far - yes I am aware about splitting or chunking the data, but interested in a good write-up of techniques and pros/cons of each with examples. Eg: Chunking sentences vs. paragraphs, providing context around the embedding result, asking GPT to generate questions to chunks and embedding that instead, combining interaction data (eg: purchases or clicks after search queries) with actual content data before embedding, embedding attributes around data, and so on.

How many tokens per second are you getting with your setup?

Paul English (founder of Kayak, exited at $2 billion) said this in response to the exact comment you are making: "Kayak was just a thin layer above ITA, and ITA sold for $700m and Kayak sold for $2b. So don’t dismiss thin layers. The ultimate power is the app and UX."[0]

I would also add "distribution" to the "app and UX" part, but you can certainly build a valuable business upon an API "everyone else has access to" - plenty of companies out there that do that.

[0]https://twitter.com/englishpaulm/status/1623701758781558784

Thanks (to both you and the parent) for sharing these details. So is it fair to say the following:

1. Fine-tuning bakes the knowledge into the model, but getting the "source" of an answer to a specific question becomes cagey and it is unclear if the answer is accurate or just a hallucination.

2. Therefore vector databases, which can provide context to the LLM before it answers, can solve this "citation" problem, BUT:

3. We then have limits because of the context window of the LLM to begin with.

Is that a fair understanding, or have I totally gotten this incorrect?

Edit: Or, are you saying that you both fine-tune AND also use a vector database which stores the embeddings of the dataset used to fine-tune the model?

Is this a script to set up a cloud provider box as a VPN to tunnel your traffic through, or something else?

Outline is part of Jigsaw, which is a part of Google. Is it truly private?[0][1]

[0]: https://getoutline.org/faq/ under the "Outline Brand" section

[1]: https://github.com/Jigsaw-Code/?q=outline Jigsaw's Github with the Google affiliation.

Interesting to see if/how this is different (or maybe built upon) Segment Anything[0].

For folks in the know: I often see segmentation models on video frames producing patchy results (see the DinoV2 video of the running dog, the body gets black patches randomly, so the segmentation fails for certain frames). What methods are folks using to deal with this - standard fine-tuning, or is there a way to "force" the area to be cleanly segmented (ie, add a bounding box around the class to supplement the data)?

And is it something that can be implemented in foundation models, or are we always going to have patchy zero-shot results like this on video files?

[0]https://github.com/facebookresearch/segment-anything

This is HC in 2020 claiming that 2016 was illegitimate: https://news.yahoo.com/hillary-clinton-maintains-2016-electi...

This was from a quick google search, there are probably more sources and more stories about the others you mentioned. There is no one candidate to blame here, most of them have some sort of story they concoct about why they lost and some conspiracy they attribute it to. The blame lies with anyone who is such a tribal political member (D or R) that they simply cannot accept anything non-nightmarish about the other side.

Nixon did it too in 1960 for JFK's election, even though that has now been determined to have been "fraudulent" in how most people understand the term: https://www.politico.com/magazine/story/2016/10/donald-trump...

Paying lip service to a failed process does not make the process more robust, it simply papers over the issues until it finally breaks, to worse outcomes and pain. There are enough elements that are questionable in the US election system that it is not unreasonable for people to be doubtful of it, esp. in the hyper-partisanship of today.

The solution is not to pretend to go back to an era of gentlemen's agreements, but rather harden the process so that nobody except the real cranks and lunatics could feasibly say anything. The problem is that too many people now are happy to write off entire swaths of the country as "loonies and cranks" because of their own hyper-partisanship.

This is ridiculous (as usual) - what are the alternatives folks are using? Or are there really no options like "git push heroku master" where everything just works (or at least, used to work magically).

You're right, Python's ecosystem and dependency management is a shitshow and everybody involved should be ashamed of themselves. But of course there are many "you're holding it wrong" commenters here who are in denial of this fact. It's an absolute pity that Python has become the de-facto language for public ML projects.

Nice to see Tortoise being used - I still think it's the best TTS system out there now. Generation time is slow, but quality is incredible. I wonder if the code can be optimised to speed up the generation, but I don't think the author is maintaining it any longer.[0]

[0]https://github.com/neonbjb/tortoise-tts

And why would that be bad? People actively in the drug trade are profiting off the misery of others and the degradation of society. There is a host of associated crimes that occur due to drug-use and in places where drugs and drug-abusers are prevalent.

If people in a society vote to deal with this issue, and their solution is to execute people in the drug-trade, why is that bad? Why is Rayiner's legitimate opinion being considered illegitimate by you?

Amazing work! Someone is probably already working on porting it to work as an extension on Stable Diffusion webui.