41st Edition - Last Week in AI - FIDO2 Project - Theory of Mind in LLMs

Welcome to the 41st Edition of Future Forward - the Emerging Tech & AI Newsletter!

This newsletter aims to help you stay up-to-date on the latest trends in emerging technologies and AI. Subscribe to the newsletter today and never miss a beat!

Subscribe to the newsletter here.

Here's what you can expect in each new issue of the Emerging Tech & AI Newsletter:

• A summary of the top AI news from the past week
• Primer on any emerging technology or recent advances in existing technologies (We review FIDO Alliance this week)
• A key topic in AI or Examples of how AI is being used or How it will impact the future ( We explore Theory of Mind in LLMs this week)

The field of AI is experiencing rapid and continuous progress in various areas. Some of the notable advancements and trends from the last week include:

Big Tech in AI

1. Google released technical details on its Gemini 1.5 model family.
2. Microsoft introduced a new category of Windows PCs designed for AI, Copilot+ PCs.
3. Dell Technologies Expands Dell AI Factory with NVIDIA to Turbocharge AI Adoption.
4. Google releases demo of Project Astra solving maths equations.
5. Meta’s AI research lab introduced Chameleon, a new family of ‘early-fusion token-based’ AI models.
6. Google’s A.I. Search Errors Cause a Furor Online.
7. Amazon plans to give Alexa an AI overhaul.
8. Amazon, AI startup Hugging Face pair to use Amazon chips.
9. WWDC 2024 to reportedly showcase Project Greymatter, Apple's idea of AI for the masses.

Funding & VC Landscape:

1. AI data-labeling startup Scale AI raised USD 1Bn in funding.
2. French AI startup H announced $220M in funding to develop advanced AGI that can automate complex business tasks.
3. ThinkLabs AI Raises $5M in Seed Funding.
4. Skribe.ai Raises $3.5M to Propel LegalTech Innovations.
5. Patronus AI Raises $17M in Series A Funding.
6. Superlegal Raises $5M in Seed Funding to Advance AI-Driven Legal Contract Review.
7. Coactive Systems Raises $30M in Series B.
8. WitnessAI Raises $27.5M in Series A Funding.
9. Expedera Raises $20M in Series B Funding.
10. Videra Health Raises $5.6M in Seed II Funding.
11. Daffodil Health Raises $4.6M in Seed Funding.
12. DeepL Raises $300M Investment at $2 Billion Valuation.
13. Haz Raises $1.4M in Pre-Seed Funding.
14. QA.tech Raises Additional €3M in Seed Funding

Other AI news:

1. IDEA Research releases Grounding DINO 1.5
2. OpenAI is rolling out new data analysis tools in ChatGPT.
3. Neuroscience startup BrainBridge introduced a new concept for an AI-driven robotic system capable of performing head transplants.
4. IBM will open-source its Granite AI models.
5. C4AI launched Aya 23, a family of state-of-the-art multilingual LLMs.
6. OpenAI secures multi-year deal to license content from News Corp.
7. AI is poised to usher in a new level of concierge services to the public.
8. AI headphones let the wearer listen to a single person in a crowd, by looking at them just once.

Liked the news summary. Subscribe to the newsletter to keep getting updates every week.

FIDO2 Project

The FIDO Alliance, short for Fast Identity Online Alliance, is an industry association formed in February 2013. Their main goal is to move us away from relying on passwords for online accounts.

FIDO works on creating standards for secure authentication that can be used across different devices and platforms. This includes things like fingerprint scanners, facial recognition, and security tokens - anything to replace passwords with a more secure and user-friendly method.

One of their recent advancements is the concept of passkeys. FIDO passkeys ditch passwords altogether, offering a faster, simpler, and far more secure way to log in to websites and apps across all your devices. Built on FIDO standards, passkeys rely on strong cryptography, unlike passwords which can be easily stolen or guessed. This means no more remembering complex passwords or dealing with frustrating phishing attempts.

FIDO2 makes logging in more secure and convenient by offering different options for authentication, both without passwords and with them.

FIDO2 combines two things:

• Web Authentication (defined by the W3C): This sets the rules for how websites and applications talk to your device to verify your identity.
• Client to Authenticator Protocols (CTAP) (developed by the FIDO Alliance): This specifies how your device interacts with different authentication methods, like your fingerprint scanner or a security key.

FIDO2 gives you flexibility in how you log in:

• No password needed: You can sign in without a password at all, using just your fingerprint, face recognition, or a PIN.
• Extra security: You can add an extra layer of security on top of your password with a security key, phone, or wearable device.

Interested in knowing more? Do let us know in the comments and we will share more about FIDO2 in a new article.

Theory of Mind in LLMs

Theory of Mind (ToM) is the ability to understand that other people have their own thoughts, feelings, and beliefs that can be different from your own. It's a complex skill that allows us to navigate social situations, empathize with others, and even deceive them.

In the context of Large Language Models (LLMs), ToM refers to their ability to process information and respond in a way that reflects an understanding of mental states. This could include things like:

• Recognizing that someone might say one thing but mean another (understanding sarcasm)
• Predicting how someone might react in a given situation (based on their beliefs and desires)
• Following a conversation that involves multiple people with different viewpoints

There's ongoing research on whether LLMs truly possess Theory of Mind. With LLMs becoming woven into our personal, professional, and social interactions, and entrusted with increasingly consequential choices, ensuring they reflect our values becomes paramount. Theory of Mind research offers a promising path forward, equipping LLMs to understand human intentions and navigate our world with greater alignment to what matters most to us.

To gauge how well large language models (LLMs) grasp human thought processes, a recent study pitted them against 1,900 human participants in a series of Theory of Mind (ToM) tests. These tests, designed to assess understanding of false beliefs, indirect requests, irony, and social faux pas, compared the performance of two LLM families: GPT and LLaMA2.

Humans seem to be born with the potential to understand other minds. But this ability needs social interaction and other experiences over many years to fully blossom. People can develop stronger or weaker abilities to understand the thoughts and feelings of others. Research suggests a close link between theory of mind development and language development in humans. A meta-analysis found a moderate to strong correlation (correlation coefficient of 0.43) between performance on theory of mind tasks and language tasks.

Studies have shown that LLMs can perform well on tests designed to assess theory of mind in humans. This is unexpected and challenges the idea that LLMs simply rely on statistical patterns in language. However, some researchers argue that LLMs might be achieving these results through clever pattern recognition and statistical analysis of the data they're trained on, rather than a genuine grasp of mental states.

There's no definitive answer yet on whether LLMs truly understand theory of mind. More research is needed to understand the mechanisms behind their performance on these tests.

References:

Original Paper on ToM as emergent capabilities of LLMs

Similar Tests.

Theory of Mind abilities of Large Language Models in Human-Robot Interaction

Disclosure: Some content in the article was written with the help of Google Gemini.

Thanks for reading. See you next week!

Let's explore the future of technology together!

Your Turn:

Did you like the content? Share with your network. Join more than 1000 people and get updates every week by subscribing to the newsletter.