I'm looking to get a car these days in EU. The government gives out subsidy for electric cars and even without it, there is simply nothing almost as good in Teslas range. EU makers are either too expensive or try to sell you shit for gold
It reminds me of the general idea of generative adversarial networks where instead of a teacher and learner, you have a one trying to learn real inputs from fake inputs while the other tries to trick the first with fakes, with both training each other and the ultimate goal is to have the latter create realistic fakes.
I'm sure it's been or is being researched, but since I'm new to LLMs, my immediate thought was having code-aware models where one tries to find and fix vulnerabilities in a codebase while another tries to exploit them where the ultimate goal is to secure a codebase.
I feel like the current meta on finetuning LLMs is random accounts at X/Twitter. Google results are littered with SEO garbage or some kind of guides that fail to work the moment you need something slightly different.
Almost no serious user - private or company - wants to slurp their private data to cloud providers. Sometimes it is ethically or contractually impossible.
We can't use LLMs at work at all right now because of IP leakage, copyright, and regulatory concerns. Hosting locally would solve one of those issues for us.
If the board is unhappy about that they are idiots and should not be board members.
Absolutely no one could have predicted Sam being removed as CEO without anyone knowledge until it happened.
But regardless a 10b investment has yields huge results for MS. MS is using openAI tech, they aren’t dependent on the openAI api or infrastructure to provide their AI in every aspect of MS products.
That 10b investment has prob paid itself back already and MS is leveraging AI faster than anyone else and has a stronger foothold on the market.
If the board can’t look past what 10b got then. I wouldn’t have faith in the board.
He was a marketing person I believe when Bill was in MSFT. To become the CEO of MSFT is a huge political and competence firewall already. Then to do the most spectacular transformation of a mega-corp is next-level. MSFT is now the leading player in AI, while before it was still fucking around with office and Windows licenses. People who are young (not saying you are), and don't remember what MSFT was before Satya, don't really get that MSFT would be like Oracle and IBM if not for Satya.
As far as I know, he actually came from an engineering background, making his career even more impressive. Despite my views on Microsoft and shareholder-oriented capitalism, he certainly seems like a brilliant and genuinely interesting guy.
The board saw cloud compute was gonna be big. They made him the king. Good bet. The whole company went all in on cloud. Now they print more money than before.
Marketing person lol. He's an engineer. The guy literally gets back into VS code sometimes to stay in touch.
There would be no way to find out if that would be true however. The person in the vat might say they are the same person but you don't know if they really are. Also what happens if you create a copy? Is that two people or is there some kind of shared consciousness? If it is, how do they communicate?
This is important work. And think about African or southeast Asian languages, they are even more screwed. We need to make sure that AI multilingual to avoid total English domination of culture.
Languages are like frameworks, they (slightly) guide your thinking. Think about the same stuff in different languages and you’ll probably get more ideas about it than in one language only.
Moreover, censorship does not cross languages. Its very hard to make a multi-lingual censor system.
Being fluent in two major language systems, its very, very glaringly obvious, what are the taboos, politically correct untruths in both languages, that seem completely invisible to the monolingual speakers.
That's the Sapir-Worth hypothesis. It's decidedly pre-Chomsky, and as Pinker calls it, it is both the most well known and accepted linguistic hypothesis and also almost certainly completely wrong.
It’s something you just know if you’re bilingual or more. It would sound threatening to monolingual speakers but this is just how it works. Languages are our thought construction algorithms.
It’s not a “Turing Completeness isn’t real” hypothesis, more like just “Rust != C”. I think this is where software nerd types predict wrong, as it sounds as if it is trying to disprove TC, which shall be futile attempt. It’s not(the former one at least).
It is obvious. Most words in different languages aren't 100% equivalent - they have large or small differences in sets of connotations. Same is true for phrases and sentences. When you translate a thought expressed in one language to another, you may get the primary, leading-order meanings across 100% right, but you'll still lose some lower-order connotations.
For a more direct analogy, I'd compare this to how LLMs process tokens, but I feel most of the community is not ready for it yet, as we're still stuck debating the validity of this comparison in the other direction...
> I have never experienced that I need to think in a specific language in order to do something better (well, I only have two choices).
I found that for me, certain thoughts "flow better" in one language, and others in another. And this makes sense, because thinking is, in large part, exploring the web of associations and connotations, going down the gradient of what "feels right"[0] - and since those annotations and connotations have different structures in different languages, so will the thoughts drift in different directions, take different paths, even if they end up in the same place.
--
[0] - The obvious parallels between one's inner voice and the workings of an LLM are left as an exercise to the reader.
A lingua franca is useful and probably inevitable. The downside is language and cultural loss. Works in translation are rarely quite as good, especially humor and wordplay. This is why various countries have "local language quota" rules for media; although derided by English speakers and HN, they're a way to keep the local language, culture and identity alive.
Apriori assuming that those things are somehow intrinsically valuable. These decisions (eg. language quotas) are made by armchair intellectuals with career/identity investments in the said language. Language evolution ignores them, but they can be annoying in the interim.
Firstly, everything is only valuable in the sense of being valuable to someone or some people; nothing is intrinsically valuable.
Secondly, everyone has a career/identity investment in a language.. their first language. The one they work in, went to school in, read in, talk to their family in, consume literature in. (I suppose HN devalues literature as well).
For monoglot anglophones struggling to understand the concept, imagine if the US declared its official language was now Standard Mandarin. There would be riots. Heck, I've seen Americans get mad at the mere use of Spanish, the country's second language.
>Firstly, everything is only valuable in the sense of being valuable to someone or some people; nothing is intrinsically valuable.
Agreed, so when tech and globalisation are pulling towards unification for practical reasons - your argument reduces to "I'm going to force others to use what I like because I don't like the decisions people are making".
I'm not arguing we should make anything official or force anything.
Yes ? If everyone suddenly started using English instead of Croatian it would be a net positive in my book. I'd probably get a lot better at casual English which is something I notice talking with non-Croatian speakers outside of work.
I'm not saying we should force adoption of English ! Tech and globalisation are pushing in that direction in the "west". People I see pushing back the most are "intellectual elites" invested in the language, presenting their value judgements as objective arguments.
But everyone wouldn't suddenly start using English, that's not feasible. Is everyone in your family fluent in English (my family is not)? In every country, there are people who couldn't adapt and disproportionally so in vulnerable groups. Suddenly, you would be in a country where you don't know the language. Everyone would be immigrants in their own country.
Protection of national languages is as much on the agenda of politicians, including right-wing populists. In your view, are the songs, stories, books and plays in Croatian the property of the elites and not of every Croatian?
You say we wouldn't force it, but if we left it to Hollywood etc. they would force it. Leaving issues for the markets to decide freely brushes aside all the negative externalities.
Not from an aesthetic sense. I think it's really cool that we have a lot of languages. I'm personally willing to pay a high price in inconvenience to keep that coolness around, although not everyone would.
However I also don't think we will have to. Machine translation and language learning are substitute goods -- the better the former gets, the fewer people who will feel any desire to pursue the latter, because it just won't be that big of a deal to translate between X and Y anyway.
A universal second language for commerce is a fine middle ground, though.
The languages you think in affect your decision making, your creativity, how you perceive the world. If we were restricted to a single language, we’d lose as individuals and as a species.
It is a concern because presumably most people in office jobs are going to need to be able to use these tools, but I am somewhat comforted to know one language that AI systems do not understand well yet because of lack of texts. However, I think that will be short lived.
While I can speak in Portuguese without much issues (except being hard for them to stick to European Portuguese), I've nooticed that sometimes it uses a clear translation of an English expression that does not feel natural in Portuguese at all.
A long time ago I had an idea after leaving DEFCON about making a standard gizmo, something like a lapel pin that connects via whatever to whatever supercomputer someone is carying in their pocket and that thing presents an obvious color code. Lets say 4 quadrants with a circle in the middle. I think ideally it would be like epaper with maybe a backlight option. Non intrusive but could be vivid. This could let people know in polite society if they did not want to be recorded, were recording, how they were recording, etc. Obviously this doesn't stop anyone from doing whatever, but you can monitor signals from people and if someone is doing something they say they aren't that can be shared. You can also transmit information like PGP public keys via ultrasonic handshake or whatever.
I mean, day to day I could see never needing it but going into the office, taking the train, going to court, going to a concert, there are a lot of ways something like that could be very useful and if adopted at any sort of scale could error check and blacklist bad actors.
I've been working on things that basically do that for a decade. I mean I didn't really think of this specific application initially but I did kind of.... rotating parabolic ultrasonic arrays something something. I'm looking for partners and investors!
I recently threw up a hackaday for my development cyberdeck and am sifting through all the stuff from over the years to finally get this to market. With the new CloudFlare AI workers, a whole bunch of the infra I used to have to maintain is a moot point so I'm looking to really hit it hard in Q1 2024.
Yeah. I'm working on that. This isn't really a pitch and you probably wouldn't want my cyberdeck. It's my dev workstation and yeah, no one wants that. I mostly mentioned it just because I'm getting to a phase where I am interested to get eyes on the weirder parts. Honestly, the concept itself is kind of a filter. If you look at it and get why you would want it, then we should probably talk, you're likely doing some weird stuff yourself!
The real pitch is working towards seamless natural user interface for streaming composite spatial data. The easiest thing to explain that people will get is architectural pre-visualization. You've got plans and lot lines, input plans, it generates a structure, that's a known solution, easy peasy. That is layered over GIS data from whatever source. Construction is underway, I can do drone surveying and flightpath automation to do photogrammatry or NeRF or whatever to build and overlay the model from the scan. Simulations are easy for erosion, or looking at the light at different times of year. I can drop ship you a 3D printed architectural model or some sort of widget. If you point your phone at that widget, you can interact directly with the information. Standard AR fare, you can go onsite and see the AR stuff. Builders can take scans of things in production and combine them. Yadda yadda. I have a company that is selling basically that. But to get to there, what I'm after is real time streaming composites with effortless and inexpensive mcguffens.
Composing things for a real time stream to be delivered for people to look at needs to have sort of LOD fallback. Streaming compressed point clouds that can be altered in real time, cameras that can be controlled or tabbed through, pulling video from multiple sources and muxing all of that into a cohesive digital product attached to a physical / inexpensive medium that a person can interact with. To build these things I've also had a focus on audio reactivity. Composition based on the sounds produced in the real world and in the digital world. Say 20 people have the same mcguffin, they can all fiddle with parts in place on their desktop as an AR or VR experience, or they can just open up a browser and mess with the world like a traditional video game. Or there can be a stream of 4k to Youtube or Twitch or whatever and that can be synced up to augment.
This is a long term study on what I don't like about modern computing and what I want to see. The stuff that makes money is high throughput distributed GPU compute and GIS mapping. These days people also want LLMs in everything so I've got the stack to produce that and it pays the bills.
The parts are really starting to fall into place though and I expect I'll have my "killer demo" by the end of the year. The problem is with this really is that I can't point to something and say "it's like that" because it doesn't exist and no one else is trying to do anything similar as far as I can tell.
Ok, now I understand. There was a European consortium working on this a few years ago for product development in the automotive and aviation industry. I think Barco had the lead on that.
I'll do some digging on the euro consortium, it wouldn't happen to be AliceVision would it? That's more a euro university group, but I could see them being related.