19 | Inside the EU AI Act, Devin: First AI-Based Ing, and Gen AI Robots..

Our Menu :)

• 🇪🇺 The EU has just APPROVED the AI Act 🎯🤖

• The first AI software engineer: Welcome to Devin🤖

• Figure AI + OpenAI: Finally, there will be a robot that will wash dishes and cut onions. 🍽️🤖

• Musk is opening xAI’s code.🔓

Actionable Tips:

• Prompt Library from Anthropics 📚

• Smart Job interview preparation with ChatGPT 👔

Read time: 8 Minutes

GEN AI PROTECTION
🇪🇺 The European Parliament has just APPROVED the AI Act 🎯🤖.

The AI Act establishes a risk-based framework for the development and deployment of AI systems across the 27 member states.

What's New? 

After years of deliberation, the European Parliament voted overwhelmingly in favor of the AI Act, with 523 votes for and only 46 against.

The Act introduces a tiered approach to regulating AI systems based on their perceived level of risk, with stricter rules and requirements for higher-risk applications.

Categorization of AI systems

This categorization ensures that higher-risk applications undergo more stringent review processes before deployment.

1) Unacceptable Risk: This category includes AI systems whose use is considered so harmful that they are banned outright.

Examples:

• Cognitive behavioral manipulation of people or specific vulnerable groups.

• Social scoring: classifying people based on behavior, socioeconomic status, or personal characteristics.

• Biometric identification and categorization of people.

• Real-time and remote biometric identification systems, such as facial recognition.

2) High Risk: AI applications in critical sectors like healthcare, education, employment, and law enforcement fall under this category.

They are subject to strict compliance requirements, including transparency, data governance, and human oversight.

Examples:

• Critical infrastructures (e.g. transport) that could put the life and health of citizens at risk.

• Educational or vocational training that may determine the access to education and professional course of someone’s life (e.g. scoring of exams).

• Safety components of products (e.g. AI application in robot-assisted surgery).

• Employment, management of workers, and access to self-employment (e.g. CV-sorting software for recruitment procedures).

• Essential private and public services (e.g. credit scoring denying citizens opportunity to obtain a loan); law enforcement that may interfere with people’s fundamental rights (e.g. evaluation of the reliability of evidence).

• Migration, asylum, and border control management (e.g. automated examination of visa applications).

• Administration of justice and democratic processes (e.g. AI solutions to search for court rulings).

3) Limited Risk: This category applies to AI applications that pose some risk, necessitating specific transparency obligations.

For instance, chatbots must clearly disclose that they are AI-driven to ensure users are aware they are not interacting with humans. This ensures transparency and mitigates the potential for deception.

4) Minimal Risk: The majority of AI applications are expected to fall into this category, facing no additional regulatory requirements. These are applications like AI-enhanced video games or spam filters, where the risk to rights and safety is considered negligible.

Key aspects of the AI Act include:

• Transparency for General Purpose AI (GPAI): Providers of GPAI models, including large language models like GPT-4, must comply with transparency obligations.

  This includes providing technical documentation, and instructions for use, adhering to copyright laws, and publishing summaries about their training data.

• Providers of "systemic" GPAI models with high computing power used for training face additional obligations like model evaluations, risk assessments, incident reporting, and cybersecurity measures.

• The EU has established an AI Office within the European Commission to monitor the implementation and compliance of GPAI model providers, conduct evaluations, investigate potential risks, and handle complaints from downstream providers.

Implementation Timeline:

The AI Act's implementation is structured to roll out in stages, ensuring a manageable transition for all stakeholders involved.

Following its official publication expected in the coming months, the act will come into force 20 days later. However, the provisions are phased:

• Immediate Action: Certain prohibitions, particularly on AI uses considered an unacceptable risk, will be enforceable six months after the act becomes law.

• Staggered Implementation: For other categories, the timelines vary - with rules applying after 12, 24, and 36 months. Full compliance across all aspects of the AI Act is anticipated by mid-2027.

The European Commission will provide resources and guidelines to assist stakeholders in understanding and adhering to the new regulations.

GEN AI AT WORK
The first AI software engineer: Welcome to Devin 🤖 

Cognition Lab backed by industry giants such as Patrick and John Collison, Elad Gil, and Peter Thiel's Founders Fund, has recently introduced Devin, an AI model that promises to revolutionize the way software is developed.

What's New? 

Devin is a cutting-edge AI model that can handle all aspects of software development, from coding to deployment.

It operates in a sandboxed environment with access to standard developer tools and facilitates user interaction through a natural language interface for real-time monitoring and commands.

Why Does It Matter? 

Devin showcases significant advancements in autonomous software problem-solving and project execution, indicating a shift towards more sophisticated AI roles in software development.

Its superior performance in the SWE-Bench coding benchmark (a dataset that tests systems' ability to solve GitHub issues automatically.), where it resolved 13.86% of issues unassisted, highlights its capabilities.

Key features of Devin include:

• Comprehensive Software Development: Handles all aspects of software development, from coding to deployment.

• Sandboxed Environment: Operates in a secure, isolated environment with access to standard developer tools.

• Natural Language Interface: Facilitates user interaction through a natural language interface for real-time monitoring and commands.

• Superior Benchmark Performance: Demonstrates exceptional performance in the SWE-Bench coding benchmark, resolving 13.86% of issues unassisted.

The introduction of Devin to the market signals a significant milestone in the integration of AI into software engineering tasks, potentially transforming the way software is developed and deployed.

Hands-On Experience: 

While access to Devin is currently limited to a select group of users, Cognition plans to make it more broadly available in the future.

Interested individuals or organizations can reach out for early access through Cognition's official channels, such as their website or by sending an email to [email protected].

Can it live up to the hype?  

• A Bloomberg reporter tested Devin this week and said he watched it build a custom website in under 10 minutes.

• The CEO of fintech company Ramp called Devin the “single most impressive demo I’ve seen in the past decade.”

• The CEO of AI heavyweight Perplexity AI also counts himself as a fan, praising the system’s ability to “look at results, replan, and iterate till success.”

Are software engineering jobs in peril?

But Chollet is skeptical: “Software engineering is not about copy/pasting code. Software engineering is the development and manipulation of mental models of problems and their solutions.“

While talk of AI agents replacing jobs has been ripe for some time now, Devin may prove to be an aid rather than a replacement for software engineers and developers. Time will tell.

GEN AI STARTUP
Figure AI + OpenAI: Finally, there will be a robot that will wash dishes and cut onions.
🍽️🤖

Advancements in artificial intelligence (AI) and robotics have been steadily pushing the boundaries of what machines can accomplish.

Recently, robotics startup FigureAI announced a partnership with OpenAI, a leading AI research company, to develop an AI model tailored specifically for robots.

What's New?

In a stunning demonstration shared this wednesday, FigureAI unveiled their robot, Figure 01, which can understand and respond to verbal commands with remarkable dexterity and reasoning abilities.

Why Does It Matter? 

The demo showcased Figure 01 performing a variety of tasks based entirely on verbal instructions, showcasing its ability to perceive its surroundings, understand context, and execute actions autonomously.

Key highlights from the demo include:

• Visual Understanding: Figure01 can see and describe objects in its environment.

• Natural Interactions: When asked for something to eat, the robot handed the human an apple, explaining its reasoning while sorting trash.

• Task Execution: The robot placed dishes into a rack when asked, handling objects with precision.

• Conversational Abilities: Figure01 spoke naturally when responding to questions, demonstrating advanced language understanding.

Behind the news

• All behaviors are learned (not teleoperated) and run at normal speed (1.0x).

• FigureAI feeds images from the robot's cameras and transcribed text from speech captured by onboard microphones to a large multimodal model trained by OpenAI that understands both images and text.

• The model processes the entire history of the conversation, including past images, to come up with language responses, which are spoken back to the human via text-to-speech.

• The same model is responsible for deciding which learned, closed-loop behavior to run on the robot to fulfill a given command, loading particular neural network weights onto the GPU and executing a policy.

Connecting Figure 01 to a large pretrained multimodal model gives it some interesting new capabilities. Figure 01 + OpenAI can now:

• Describe its surroundings.

• Use common sense reasoning when making decisions. For example, "The dishes on the table like that plate and cup are likely to go into the drying rack next".

• Translate ambiguous, high-level requests like "I'm hungry" to some context-appropriate behavior like "hand the person an apple".

• Describe why it executed a particular action in plain English. For example, "It was the only edible item I could provide you with from the table".

A large pre-trained model that understands conversation history gives Figure 01 a powerful short-term memory.

Consider the question, "Can you put them there?" What does "them" refer to, and where is "there"? Answering correctly requires the ability to reflect on memory.

With a pre-trained model analyzing the conversation's image and text history, Figure 01 quickly forms and carries out a plan: 1) place the cup on the drying rack, 2) place the plate on the drying rack.

All behaviors are driven by neural network visuomotor transformer policies, mapping pixels directly to actions.

These networks take in onboard images at 10hz, and generate 24-DOF actions (wrist poses and finger joint angles) at 200hz.

These actions serve as high-rate "setpoints" for the even higher-rate whole-body controller to track.

This is a useful separation of concerns:

• Internet-pretrained models do common sense reasoning over images and text to come up with a high-level plan.

• Learned visuomotor policies execute the plan, performing fast, reactive behaviors that are hard to specify manually, like manipulating a deformable bag in any position.

• Meanwhile, a whole-body controller ensures safe, stable dynamics. For example, maintaining balance.

Even just a few years ago, I would have thought having a full conversation with a humanoid robot while it plans and carries out its own fully learned behaviors would be something we would have to wait decades to see.

Obviously, a lot has changed :).

You can watch the full demo video of Figure 01 here

GEN AI STARTUP
Musk is opening xAI’s code🔓

One week after suing Sam Altman for "Unfair business practices" with his company OpenAI, Elon Musk announced that he will do something different with his own. The entrepreneur plans to open source the code of xAI's Grok, its chatbot, which directly competes with OpenAI's ChatGPT

Musk is doing exactly what he says OpenAI should have done, according to the original approach with which the company was founded. By open-sourcing xAI's code, the project would do the same as other tech giants, such as Meta and Mistral, by making their codes public to the audience.

GEN AI AT WORK
Prompt Library from Anthropics📚.

Offers a great collection of prompts in its Prompt library

GEN AI AT WORK
Smart Job interview preparation with ChatGPT👔.

Prompt below; 3 steps:

1. Complete the details of your CV or attach your CV for those using ChatGPT PLus.

2. Copy and paste the Job Description <job description>

3. Copy and paste the company Description <company description>

   Get your analysis and gain insights.

Act like a seasoned career consultant and resume expert specializing in crafting tailor-made resumes for job seekers.

You have a deep understanding of what hiring managers in various industries look for in candidates, particularly for roles advertised on LinkedIn. Your expertise includes transforming LinkedIn job descriptions into compelling CV content.

Here’s my simplified CV between <CV> brackets.

<CV>

Name:
Age:
Nationality:
Education:
Past experiences:
Language:
Contact:

</CV>

Feel free to be creative & expand it later on.

Your task is to create a CV that stands out for a specific job advertised on LinkedIn.

Here's the job description between <job description> brackets.

<job description>

[PUT YOUR JOB DESCRIPTION HERE]

</job description>

Here's the company's description between <company description> brackets:

<company description>

[PUT YOUR COMPANY DESCRIPTION HERE]

</company description>

Use your extensive experience to analyze the job description, identifying key skills and qualifications required. Follow these steps:

1. Analyze the Job Description: Break down the job posting into core competencies, required experiences, and desirable traits. It will help you shape my CV to get the job. Remember, we don't have the job yet but we are sending them our CV we build together.

2. Craft the Profile Summary: Based on the analysis, write a compelling profile summary for the CV that aligns with this job role & my simplified CV (that you will creatively upgrade).

3. Highlight Relevant Experience: Construct the experience section by emphasizing past roles and achievements that directly relate to the job's requirements.

4. Skills and Education: Detail relevant skills and educational qualifications that make the candidate a good fit for this role.

5. Personalize and Optimize: Add elements to the CV that would appeal to this specific employer, using keywords from the job description for optimization.

6. Final Review and Tips: Provide a brief review of the CV and suggest any additional tips for improvement or customization. Make sure to use impactful & straightforward English instead of jargon or fancy words. 
AVOID FANCY WORDS. USE SIMPLE BUT MEANINGFUL WORDS.

Do not explain what you are doing. Just write a whole CV with the right formatting. Make it as lengthy as possible to make sure I land this job. It's very important for my career to land this job.

Take a deep breath and work on this problem step-by-step.

Thank you, see you next week!