Dogfooding generally consists in using your own product, to make sure you feel the user’s pain points.

Anthropic is probably doing it unwantingly, as their job postings inbox is so full of AI-written applications that they had to introduce the following policy:

While we encourage people to use AI systems during their role to help them work faster and more effectively, please do not use AI assistants during the application process. We want to understand your personal interest in Anthropic without mediation through an AI system, and we also want to evaluate your non-AI-assisted communication skills. Please indicate ‘Yes’ if you have read and agree.

— Anthropic Job Application Form (via Simon Willison)

Recentemente o Primeiro Ministro anunciou na WebSummit uma contratação directa para o desenvolvimento de um LLM Português. Falou no IST e na Nova, omitindo o resto do consórcio em IA Responsável que tem trabalhado em vários aspectos relevantes (fairness, sustentabilidade ambiental e fiabilidade).

Curiosamente, um grupo de investigação do meu departamento tem já feito trabalho na área, tendo lançado dois modelos (Albertina e Gervásio) em Português Europeu. Deu agora uma entrevista muito educativa ao Dinheiro Vivo:

Por exemplo, um banco quer apenas ter um assistente virtual para os seus clientes, que fale acerca de depósitos, levantamentos, etc. Não vai querer que o seu chatbot faça tradução automática, sumarização, dê a biografia do Friedrich Nietzsche e faça piadas.

Um LLM não é um chatbot. Um LLM é, numa analogia que as pessoas compreendem, uma espécie de um motor e a partir de um motor nós podemos fazer diferentes modelos de carros. O LLM é aquilo sobre o qual se pode desenvolver diferentes aplicações, uma das quais é o chatbot, outra, por exemplo, a tradução automática, ou o diagnóstico médico, etc.

Então, a nossa proposta nesse artigo de opinião, que saiu no Público em fevereiro de 2023, é que o que precisamos de uma IA aberta e de desenvolvimento de LLMs em código aberto, licença aberta e distribuição aberta para que outros atores e outras organizações, seja da investigação, seja da administração pública, seja do setor da inovação, possam eles próprios construir as suas propostas de valor e tirar partido desses LLMs sem estarem dependentes do fornecimento desses serviços, das big techs. Portanto, quanto mais houver uma oferta cada vez mais variada, mais se reduz o risco de dependência de um pequeno oligopólico que nos fornece esses serviços.

— António Branco @ Dinheiro Vivo

I’ve been following Gwern.net for a long time, and I was really curious when I found out this interview that kept him anonymous.

How does an hearing-impaired, introvert American grow up to be a polymath and a scholar?

The wikipedia editor, now full-time writer without a salary
got a lot farther by writing a lot and harder,
by being a lot smarter, by going down rabbitrabbitholesholes
by fourteen, they placed him in charge of wikipedia edits
🎶

Because I’ve always seen reading time as such an incredibly personal thing, I’ve never once paid any heed to these widgets. In fact, I’ve been slightly perturbed a service would presume to know how quickly I could read an article.

I’ve always viewed any service that sticks a “reading time” widget on its articles as the literary equivalent of fast-food: you’re not here for quality, but for expediency.

Personally, I think they devalue a text more than they add to it.

Estimated Reading Time Widgets — Jim Nielsen

I’m with Nielsen on this one. It makes no sense to suggest an (average) estimated reading time, unless you are trying to convince your audience that this is not a long piece, and you can read it in a minute or two. For personal blogs like this one, I really don’t care if people read it or not. But if you are writing one of those linked-in-self-promotion-bullshit-crypto-AI-latest-fad types of posts, then your goal is to get the reader to take its time. And in those cases, you can bluntly lie and always say it’s one minute.

My guess is that this trend comes in response to instagram/reels/shorts/tiktoks where people do not want to spend any more time on a particular content than they need. Long-form reading is dying, especially on the internet.

Other than teaching how debilitating this is to the younger generation, I hold a very pessimistic view for the next generation.

During my current visit to UCL’s SOLAR group, I attended this week’s SSE Seminar on the challenges posed by the AI Act, presented by Paolo Falcarin.

The European Union Artificial Intelligence act is EU’s first attempt of regulating AI products and services. It defines different requirements based on the risk level of the application (ranges between high, medium and low).

In particular, high-risk usage (healthcare, toys, security, …) have stricter requirements. To begin with, they need to be registered in an European database, and frequently updated as the implementation or requirements change. Furthermore, the service or product should be documented, should be traceable, transparent, secure and overridable by humans. Despite these requirements, there is no clear definition, or path forward on how these properties can be ensured, especially when applications are closed-source and frequently trained and updated. One of the challenges we addressed in CAMELOT was how to build interprable Machine Learning models. We explored the use of Genetic Programming and Domain-Specific Languages to create inherently interpretable models. My research team is continuing exploring that possibility.

General-Purpose models have a special categorization within the AI act: Providers of general-purpose models (think OpenAI or Google) must provide a good understanding of the capabilities and limitations, comply with European copyright law, and provide a sufficiently detailed summary about the content used for training of the general-purpose model, following a given template.

Despite all the effort in understanding the potential of deep neural networks, and generative models in particular, it is not clear year what exactly are the capabilities or limitations of what they can produce. Without a clear standard of what is expected, organizations might be able to completely ignore this requirement.

As an example, take the Gandalf AI game, in which you can (easily) trick the LLM into telling you the password of the next level, even though it was instructed not to. Even with the aditional blocks introduced by each level, it is still easy to win the game. This is the state of the art in protecting LLMs from producing a known output. On a larger scale, an earlier version of Microsoft’s AI-powered Bing also generated output that went against the policies it was purposely trained against.

As for the copyright compliance, this goes in direct conflict with what OpenAI is defending. In fact, using copyrighted materials for free in the context of learning is allowed by European law. As such, it is not clear what this article entails in practice. My guess is that this is going to require a reform on the copyright law, possible to distinguish human versus automated learning. Otherwise, models this efficient may never (legally) exist again. This is something the law and computer science communities should debate before politicians take the initiative.

Finally, general-purpose AI with systemic risks (probably all of Large Language Models) have stricter restrictions: they need to evaluate models based on standardised protocols and tools, documenting adversarial testing of the model. While there are good practices for evaluating models, I do not believe the community will agree on an universal metric for general-purpose AI, and different metrics will arise.

Overall, I think it is positive that the EU is trying to regulate the use of AI. Unfortunately, I think it is a lost battle, as the technology is very new and evolves at a very high pace (something that beaurocracy might slow down). I defend that the EU should invest more on the evaluation and monitoring of AI, maybe more than on its development. After all, we cannot compete with the likes of NVIDIA, OpenAI/Microsoft, Google, Apple or Amazon, since they aquire all the relevant hardware before Europe does. Not even Intel or TSMC investing in Europe will allow us to beat US companies (or adjacent universities) in AI research. However, we can beat them in studying the societal impact of AI.