NTEU

Neural Translation for EU.
Developing the largest direct language to language neural engine farm - and 15M segments of training data
2018-EU-IA-0051

[Pangeanic @Pangeanic]

Reliability in an LLM depends on curation, not volume. Moving from Fine-Tuning to Red Teaming requires traceable Data Pipelines, RLHF alignment, and hallucination mitigation under the EU AI Act. The technical infrastructure behind AI Data Operations:
https://t.co/hGkUF776IA

[Pangeanic @Pangeanic]

Multilingual AI rarely fails everywhere at once. It fails locally.

Our new article explains language coverage, tokenization, annotation, evaluation and Local Quality Collapse.

https://t.co/YvjyQQyOul

#MultilingualAI #AIData

[NTEU @NTEU3]

@PipoVargas13 @jorgecarrion21 Yo combino @Alibaba_Qwen con @OpenAI y los resultados son bastante óptimos . Diferentes lógicas y entrenamientos pillan distintos

[NTEU @NTEU3]

@PipoVargas13 @jorgecarrion21 Yo combino @Alibaba_Qwen con @OpenAI y los resultados son bastante óptimos . Diferentes lógicas y entrenamientos pillan distintos

[NTEU @NTEU3]

@PipoVargas13 @jorgecarrion21 Yo combino @Alibaba_Qwen con @OpenAI y los resultados son bastante óptimos . Diferentes lógicas y entrenamientos pillan distintos ángulos y detalles

[NTEU @NTEU3]

@jorgecarrion21 @GeminiApp peta mucho

[NTEU @NTEU3]

Our https://t.co/QKrXpW7BDI project may be over but neural machine translation is not and in fact, it is a very good case of “small task-specific models” @Gartner_inc talks about. More on the subject : https://t.co/LZpk1WxC35

[NTEU @NTEU3]

@Pangeanic Wow! I wish AI will improve democratic institutions!

[Pangeanic @Pangeanic]

Experience Next-Gen Translation Technology!
Our latest video showcases Deep Adaptive AI Translation in action, achieving true human fluency from English to Korean!
See how our RAG-based system & LLM refine translations.

https://t.co/UYhcUGoIwm

#AITranslation #DeepLearning

[Yann LeCun @ylecun]

Very interesting paper: using generative AI to produce text or images emits 3 to 4 orders of magnitude *less* CO2 than doing it manually or with the help of a computer.

https://t.co/ErIPs4jCpM

[NTEU @NTEU3]

Interesting take on how #AppleTranslate could compete with the rest of the pack… through #data ! https://t.co/yZYMzPpAZ3

[Pangeanic @Pangeanic]

Pangeanic builds another strategic alliance thanks to its innovative #NaturalLanguageProcessing tools🔨

👉Want to find out more? https://t.co/3sQvd7F2WE

[Yann LeCun @ylecun]

We all agree that we need to arrive at a consensus on a number of questions.
I agree with @geoffreyhinton that LLM have *some* level of understanding and that it is misleading to say they are "just statistics."
However, their understanding of the world is very superficial, in large part because they are trained purely on text.
Systems that would learn how the world works from vision would have a much deeper understanding of reality.
Second, auto-regressive LLM have very limited reasoning and planning abilities.
I do not believe we can get anywhere close to human-level AI (even cat-level AI) without
(1) learning world models from sensory inputs like video,
(2) an architecture that can reason and plan (not just auto-regress).
Now, if we have architectures that can plan, they will be *objective driven*: their planning will work by optimizing a set of objectives at inference time (not just training time).
These objectives can include guardrails that will make those system safe and subservient *even* if they end up having much better world models that humans.
Then, the problem becomes to design (or train) good objectives functions that will guarantee safety and efficiency.
It's a hard engineering problem, but not as hard as some have made it to be.

[NTEU @NTEU3]

A resource for parallel data for #machinetranslation https://t.co/XxtxQENajS

[Yann LeCun @ylecun]

The evidence is accumulating.

[NTEU @NTEU3]

A resource for parallel data for #machinetranslation https://t.co/XxtxQENajS

[Spiros Margaris @SpirosMargaris]

Universal basic income of £1,600 a month

to be trialled in two places in #England

https://t.co/wj9cHB4Es3 #fintech #futureofwork #UniversalBasicIncome #UBI @Dahabaalihussen @guardian

[The Short Straw @short_straw]

I promised a thread this weekend about OpenAI and the lawsuit I filed against them, and an explanation of what I hope to achieve here. Sorry for the length, but there's a lot going on here.

To begin with, we need to understand what “OpenAI” really is: a poorly constructed scheme operated by the Y Combinator Group to avoid the requirements of copyright law, tax law, and fair competition law in order to enrich their own insider group, the Y Combinator network.

What’s Y Combinator you ask? YC is a ‘tech accelerator’ out of San Francisco, CA. They work with tech startups to scale them up at alarming speeds (a process called Blitzscaling) and provide access to the Venture Capital money that drives these products. YC has funded over 4,000 startups since 2005. When you think of YC and blitzscaling, think of companies like Airbnb and Instacart; they captured incredible portions of market share with their VC funding and grand promises to the gig economy workers they pray on, then rapidly raise their rates and include absurd restrictions and costs until they are making a bare minimum 20% cut off of every transaction to repay these VCs. As these companies mature, they min-max to suck every penny they can out of local economies, there is nothing altruistic about them. They are very proud of this mode of operation; they refer to themselves as ‘Masters of Scale.’ It is also notable that Microsoft has invested in many YC companies.

Let’s look at why I claim OpenAI is just a poor cover for the activities the YC group:

1. Of the notable board members, investors, and business partners of OpenAI, well over 90% of them have previous financial dealings with Sam Altman and YC. Read the complaint, be horrified when you see the list.

2. Sam Altman stepped down as president of YC to become CEO of OpenAI. – Same leader

3. Board members, technology, corporate structures, and even the names are constantly passed back and forth and used freely among this group. For example YC Research, once a Research arm of YC, is now called “OpenResearch.” This is called comingling. When a corporation is formed, it has a separate legal identity from its individual owners. Comingling is generally illegal whether you are for-profit or non-profit. All the resources of the non-profit arm of OpenAI are comingled with resources of the for-profit arm. Same board members, same technology assets, and again, even the same name. OpenAI confirms all of this in a blog post.

4. OpenAI has claimed from founding that the company mission is non-profit; they have committed to ‘broad distribution’ to ‘benefit all of humanity.’ However, a close look at forming partnerships easily debunks this claim. While many developers are on a waiting list to create apps working with OpenAI’s GPT software, the business partners who received early access and are at the top of this list are overwhelmingly YC network founders and investors. Devs, while you are waiting for access right now, over 70 YC companies from the latest W23 batch are ‘AI companies.’ These companies will get access to OpenAI’s GPT and API while you are still waiting. The game is rigged; by the time you get access YC will have lined their pockets and captured the majority of market share. It is the Y Combinator network, not ‘all of humanity,’ that is reaping significant financial benefits from the founding and operation of OpenAI.

5. The members of the YC network came together and pooled their resources under the banner of ‘OpenAI’ to avoid copyright protections, paying taxes, and create, for themselves, the world’s most powerful supercomputer.

So as a member of the group ‘all of humanity,’ the reported beneficiaries of OpenAI, I have filed a claim for Breach of Fiduciary Duties. Breach of Fiduciary Duties is the umbrella my complaints generally fall under. The board members of OpenAI have a legal duty to serve the mission of the non-profit in good faith, to act prudently, and to obey the law, all of which they are not doing. OpenAI has *constantly* confirmed via public messaging their mission is to “benefit all of humanity;” the only logical conclusion to these statements is that all of humanity, you, me, and people on every continent, are beneficiaries of OpenAI’s mission. *I firmly believe, and state in good faith, that every human being on the planet has standing to sue OpenAI for these egregious actions.* Further, I look forward to hearing the legal team of OpenAI explain how this is not the case. If anyone has any illusions about the altruistic intentions of OpenAI, their explanation should put these illusions to bed.

Do I think my claim will go anywhere? Hard to say. The defendants are overwhelmingly some of the richest and most connected parties in the world. They have a combined of net worth of trillions of dollars. I’m litigating the suit on their home turf in SF, CA. However, I hope some attention on my claim will make it obvious that the State Attorneys General should be litigating and correcting this situation. OpenAI has repeatedly promised its benefits to “All of Humanity” in a ‘broad and distributed fashion.’ Any SAG that isn’t bought and sold by the upper class should be filing suit on behalf of their state constituents. One of my goals here is for us all, together, to bring enough attention to this problem to drag the SAGs into this. I’m happy to do the SAG’s jobs for them until they wake from their slumber.

People will make many claims about me and my complaint. People will claim I am crazy, or I have no right to file a suit, or that OpenAI has no obligations to me despite the absurd level of virtue signaling about ‘benefitting all of humanity.’ Those people are gate-keepers, and nothing they say stopped me from filing my suit, and nothing they say will convince me not to make my attempt to keep OpenAI accountable. They will tell you that everything they have done here is fair and above-board, they will tell you there will be plenty of jobs and opportunity left after they have sucked the financial landscape dry. They will tell you I’m crazy, and that if you even think of challenging them, you are crazy too. You know better; if we give an inch on this, they will take miles. Be loud. Protect your work. Protect your future. Don’t for one second think you are crazy for being concerned with any of this, you’re not.

[François Chollet @fchollet]

To be clear, at this time and for the foreseeable future, there does not exist any AI model or technique that could represent an extinction risk for humanity. Not even in nascent form, and not even if you extrapolate capabilities far into the future via scaling laws.