AI Weekly Digest #35: SQLCoder beats GPT-4, and EU AI Act Leak & Adoption Timeline

Hello, tech enthusiasts! This is Wassim Jouini and Welcome to my AI newsletter, where I bring you the latest advancements in Artificial Intelligence without the unnecessary hype.

You can find me on LinkedIn, Twitter and Medium! Let’s connect!

Now let's dive into this week's news and explore the practical applications of AI across various sectors.

Main Headlines

Here are the three main trends to keep in mind if you are working in AI today.

#1 OpenAI New Models Released

Last week, I reported on OpenAI's update regarding its new, better, and more affordable models. Its latest GPT-3.5-turbo model was finally released a couple of days ago. In a nutshell, it features a 16k context, a 50% price reduction, higher accuracy, and a fix for an encoding bug.

You can read everything you need to know about this release here (GPT-3.5, GPT-4 and Embeddings releases).

#2 EU AI ACT Leak & Adoption Timeline

The European Union has taken a major step towards regulating Artificial Intelligence (AI) with the unanimous approval of the AI Act by the ambassadors of the 27 EU countries.

Next? The next steps include

• The adoption of the AI rulebook by the European Parliament’s Internal Market and Civil Liberties Committees on February 13, followed by a plenary vote scheduled for April 10–11 2024.

• The Act will officially enter into force 20 days after its publication in the official journal, with a phased application of its provisions starting six months thereafter for prohibited practices and extending to one year for AI model obligations.

• Most rules will be fully applicable after two years, with some exceptions.

Read everything you need to know about the EU AI ACT, the leaked final draft, as well as its adoption timeline.

#3 Elon Musk’s Neuralink: First Human Trial

Elon Musk's Neuralink has initiated its first human trial for a brain-computer interface device, leading to various reactions and discussions among researchers in the field.

“This study involves placing a small, cosmetically invisible implant in a part of the brain that plans movements. The device is designed to interpret a person’s neural activity, so they can operate a computer or smartphone by simply intending to move – no wires or physical movement are required.” (Source: Neuralink website)

In a nutshell:

• Neuralink, developped a brain-reading device aiming to allow control of devices through thought, raising excitement and concerns about safety and transparency among neurotechnology researchers.

• The Neuralink chip, which is fully implanted and wireless, distinguishes itself by targeting individual neurons with 1024 recording sites, potentially offering higher bandwidth for brain-machine communication compared to other BCIs.

• Scientists express concerns over the lack of detailed information about the trial, including its registration and protocol, emphasizing the importance of transparency for volunteer safety and the advancement of BCI technology.

To be continued…

#4 Multi-Modal LLMs - A Comprehensive Survey

You might already be using OpenAI APIs to power, or supercharge, your processes and products. If so, you are probably aware of its limitations, such as access to other modalities like images, as well as privacy and GDPR risks, to name a few.

Many of us have made the choice to rely more and more on multimodal models, with a focus on open-source models, to mitigate privacy concerns.

If you are exploring this topic as well, I’d recommend reading this recent survey on multimodal large language models (LLMs).

For those who haven't explored this area yet, it's worth noting that it offers significant potential for enhancing data extraction and automation.

#5 Open Source SQLCoder-70B beats GPT-4

This week, Meta released its new foundation model, CodeLlama-70B. This model is particularly suited for fine-tuning coding models.

Defog relied on its hand-crafted SQL dataset to fine-tune CodeLlama and released SQLCoder-70B, the first open-source model to outperform GPT-4.

The result is an impressive model that excels across a wide range of SQL tasks.

Read more about the results of SQLCoder-70B and how to test it here.

This is it for Today!

Until next time, this is Wassim Jouini, signing off. See you in the next edition!

Have a great week and may AI always be on your side!