Leandro Duarte

Committed to building cool tech and launching impactful startups.

[Leandro Duarte @ldr0__7]

Last week I had the opportunity to pitch my startup @HashGameStore and showcase all the development my team has made during this last 6 months.

Reach me out if you want to share ideas about this project🚀🚀 #startup #Web3 #gaming

[Gonçalo Canhoto 🇵🇹 @goncalo_canhoto]

Ontem este tweet teve alguma tração, então meti o Fable a cozinhar de lado passado 1h de postar. Com algumas iterações, criei a Cesta

A https://t.co/ywBbszBUI4 é um MCP Connector que liga o ChatGPT, Claude, (e até clis/harnesses como Cursor / Claude Code) ao Continente Online e Pingo Doce

Features:
-Procura preços reais (com promoções) dos produtos do Continente/Pingo Doce
-Adiciona ao Carrinho (apenas Continente Online)
-Sabe quantias mínimas de venda do Continente
-Não precisas de conta, nem de password, nem de instalar nada. Os produtos entram no carrinho pela sessão que já tens aberta no Continente. Nunca vemos credenciais. Não há base de dados: a cesta viaja dentro do próprio link.

Coisas interessantes que podes fazer:
-Criar carrinhos de compra com Budgets
-Dar uma receita e receber uma lista de compras com quantidades certas
-Tirar foto ao frigorífico e dizer para encomendar o que falta
-Foto de uma receita -> carrinho
-Restrições alimentares: sem glúten, sem lactose, vegetariano a lista sai já filtrada
-Comparar packs dentro do mesmo produto

Experimentem!

PS: Not affiliated com Continente/Pingo Doce, é provável que isto vá abaixo soon
PS 2: Isto foi criado em umas horas, obviamente que vai ter bastantes bugs
PS 3: Provavelmente vai rebentar / estar lento / IP blocked

[gio @GiorgioPatrini]

"it's been a whole 13 days since someone dropped a >100B open weights model" by @Smol_AI

[Google.org @Googleorg]

Today we announced $25M in funding to organizations equipping educators and students with #AI skills, enhancing learning experiences and preparing the students for the workforce of the future. Learn more ↓

https://t.co/cxvevfdKvJ

[Andrej Karpathy @karpathy]

SQL injection-like attack on LLMs with special tokens

The decision by LLM tokenizers to parse special tokens in the input string (<s>, <|endoftext|>, etc.), while convenient looking, leads to footguns at best and LLM security vulnerabilities at worst, equivalent to SQL injection attacks.

!!! User input strings are untrusted data !!!

In SQL injection you can pwn bad code with e.g. the DROP TABLE attack. In LLMs we'll get the same issue, where bad code (very easy to mess up with current Tokenizer APIs and their defaults) will parse input string's special token descriptors as actual special tokens, mess up the input representations and drive the LLM out of distribution of chat templates.

Example with the current huggingface Llama 3 tokenizer defaults:
Two unintuitive things are happening at the same time:
1. The <|begin_of_text|> token (128000) was added to the front of the sequence.
2. The <|end_of_text|> token (128001) was parsed out of our string and the special token was inserted. Our text (which could have come from a user) is now possibly messing with the token protocol and taking the LLM out of distribution with undefined outcomes.

I recommend always tokenizing with two additional flags, disabling (1) with add_special_tokens=False and (2) with split_special_tokens=True, and adding the special tokens yourself in code. Both of these options are I think a bit confusingly named. For the chat model, I think you can also use the Chat Templates apply_chat_template.

With this we get something that looks more correct, and we see that <|end_of_text|> is now treated as any other string sequence, and is broken up by the underlying BPE tokenizer as any other string would be:
TLDR imo calls to encode/decode should never handle special tokens by parsing strings, I would deprecate this functionality entirely and forever. These should only be added explicitly and programmatically by separate code paths. In tiktoken, e.g. always use encode_ordinary. In huggingface, be safer with the flags above. At the very least, be aware of the issue and always visualize your tokens and test your code. I feel like this stuff is so subtle and poorly documented that I'd expect somewhere around 50% of the code out there to have bugs related to this issue right now.

Even ChatGPT does something weird here. At best it just deletes the tokens, at worst this is confusing the LLM in an undefined way, I don't really know happens under the hood, but ChatGPT can't repeat the string "<|endoftext|>" back to me:

Be careful out there.

[Google DeepMind @GoogleDeepMind]

What's the difference between chatbots and agents? 🤖

How could AI help supercharge our creativity? 🎨

And what could life be like after AGI is achieved? 🌌

Starting 🔜: season 3 of the Google DeepMind podcast - hosted by @FryRsquared.

Subscribe now ↓
https://t.co/wJ04GB6kfM

[Ahmad Al-Dahle @Ahmad_Al_Dahle]

We just announced a new AI accelerator with @HuggingFace & @Scaleway to support European startups. Enabling more innovation with open technologies is what moves our industry forward and I'm excited to see what these startups build with open source models.

Apply here: https://t.co/sYMrMtplUP

[SSI Inc. @ssi]

Superintelligence is within reach.

Building safe superintelligence (SSI) is the most important technical problem of our​​ time.

We've started the world’s first straight-shot SSI lab, with one goal and one product: a safe superintelligence.

It’s called Safe Superintelligence Inc.

SSI is our mission, our name, and our entire product roadmap, because it is our sole focus. Our team, investors, and business model are all aligned to achieve SSI.

We approach safety and capabilities in tandem, as technical problems to be solved through revolutionary engineering and scientific breakthroughs. We plan to advance capabilities as fast as possible while making sure our safety always remains ahead.

This way, we can scale in peace.

Our singular focus means no distraction by management overhead or product cycles, and our business model means safety, security, and progress are all insulated from short-term commercial pressures.

We are an American company with offices in Palo Alto and Tel Aviv, where we have deep roots and the ability to recruit top technical talent.

We are assembling a lean, cracked team of the world’s best engineers and researchers dedicated to focusing on SSI and nothing else.

If that’s you, we offer an opportunity to do your life’s work and help solve the most important technical challenge of our age.

Now is the time. Join us.

Ilya Sutskever, Daniel Gross, Daniel Levy
June 19, 2024

[Leandro Duarte @ldr0__7]

@bancadadeleao @sporttvportugal @tvi pff troquem de comentadores

[Google DeepMind @GoogleDeepMind]

With @Harvard, we built a ‘virtual rodent’ powered by AI to help us better understand how the brain controls movement. 🧠

With deep RL, it learned to operate a biomechanically accurate rat model - allowing us to compare real &amp; virtual neural activity. → https://t.co/GaToq3AWTQ

[Leandro Duarte @ldr0__7]

@rowancheung Great work man! Keep up

[Rowan Cheung @rowancheung]

3. Flappy Bird in one prompt

The model im-also-a-good-gpt2-chatbot was able to generate a full Flappy Bird game in one prompt, with PNG files provided.

And the prompt was simply: "Code Flappy Bird game in Python"
https://t.co/NT8VmmXO0Z

[Jack Clark @jackclarkSF]

Life is the stuff that happens in-between 'doing stuff on your to-do list' and sometimes you should willfully screw the to-do list and just live.

[Andrej Karpathy @karpathy]

The history of computing is repeating in an echo, except replace computers that do precise arithmetic on bytes with computers that do statistical arithmetic on tokens.

[Rowan Cheung @rowancheung]

Adobe just introduced a slew of new AI capabilities coming to Adobe Premiere Pro.

- New AI-powered tools in Premiere Pro
- A new proprietary AI model called Firefly Video
- Third-party AI model integrations with OpenAI, Runway, and Pika Labs

[Leandro Duarte @ldr0__7]

Big move for Amazon

[Rowan Cheung @rowancheung]

@steph_palazzolo @erinkwoo Amazon just added the brilliant @AndrewYNg to its board of directors.

CEO Andy Jassy also said in a letter to shareholders this week that generative AI may be the largest technology transformation ‘since the internet’.

[Pedro Magalhães @PCMagalhaes]

Foi preciso um paquistanês (Aboubakar) para encontrar a melhor definição de Portugal.

[Rowan Cheung @rowancheung]

Spotify introduced AI playlists in beta

The new feature adds to the popular 'AI DJ' and will allow users to generate personalized playlists based on written prompts.

[Elon Musk @elonmusk]

Should be available on 𝕏 next week.

Grok 2 should exceed current AI on all metrics. In training now.

[Aidan @aidancramer]

Nothing to see here, just my AI clone taking a job interview

[Stability AI @StabilityAI]

Today, we are releasing Stable Video 3D, a generative model based on Stable Video Diffusion. This new model advances the field of 3D technology, delivering greatly improved quality and multi-view.

The model is available now for commercial and non-commercial use with a Stability AI Membership.

Learn more and read the research paper here: https://t.co/uXwZalGBky (1/3)