Each week, we bring you 5 stories that resonated the most in our internal Slack channel #AI-news. We write the newsletter using various AI tools because we're an AI company and our marketing wants to move with the times too. ๐
Today you're reading the 43rd issue in a row.
#1
Altman's crypto project "Worldcoin" renamed "World" ๐: Aims to verify human identity in the age of AI ๐๏ธ๐ค.ย
Sam Altman's cryptocurrency project, originally known as "Worldcoin" ๐, has been renamed "World" with a new goal - to verify whether humans are really humans in a world increasingly controlled by artificial intelligence ๐ค. The project uses iris scans ๐๏ธ to create a "World ID", which is essentially your "human ID" ๐ that helps distinguish humans from bots. This move comes just as AI is starting to pop up everywhere, causing stress around who is truly human and who can be trusted online ๐.
World has expanded its services with a new version of its eye scanning device ๐ that allows users to schedule an appointment ๐ or go to stores ๐๏ธ to get their World ID. The goal is to build a system that ensures that humans - and not AI bots - can safely communicate and exchange resources ๐ฌ.
However, the project has run into criticism about privacy ๐ and the use of biometrics, especially in countries where salaries aren't as high ๐ธ. The testing has also raised questions about the wisdom of exchanging biometrics for cryptocurrencies, which have been years in the making. In doing so, some operators of these devices have had to contend with inconveniences and even dangerous conditions ๐ง. And what do you think about it ๐ค?
ย
#2
Nvidia vs OpenAI ๐ค: New AI model Llama-3.1-Nemotron beats GPT-4 ๐ ๏ธ๐.
Nvidia ๐ฎ has launched a new AI model "Llama-3.1-Nemotron-70B-Instruct" ๐ค that performs better than big names in the game, such as OpenAI's GPT-4 ๐ฅ. Without much of a launch event, this model appeared on the AI platform Hugging Face ๐, where it quickly gained attention for its high performance in key benchmarks ๐.
The model achieved top scores ๐ in several benchmarks and even outperformed both GPT-4 and Anthropic's Claude 3.5 in language tasks ๐. This is quite significant because Nvidia, traditionally known for its hardware ๐ป (such as GPUs), is now making great strides in AI software development ๐.
Nvidia is also focusing on making the outputs of these models more accurately meet the needs of users ๐ฏ. Businesses that want reliable and customizable AI solutions will find this model super useful ๐ก, and Nvidia is offering free access through its platform for experimentation ๐ผ.
With this move, Nvidia has "challenged" ๐ฅ established AI companies like OpenAI to a duel, signaling a shift in the competitive AI landscape ๐.
ย
#3
Startup Lightmatter raises $400 million๐: Photonic chips for AI are the future of data centers ๐๐!ย
Startup Lightmatter ๐, which makes photonic chips ๐ก for AI data centers ๐ค, has raised $400 million in a new round of Series D funding, bringing its value to $4.4 billion ๐.ย
These chips use light ๐ for faster and more energy-efficient data processing โก. With the growing demand for artificial intelligence, thanks to tools like ChatGPT ๐ฌ, data centers need better infrastructure, and Lightmatter's chips promise to boost AI performance while reducing power consumption ๐ฑ.ย
The company plans to use the money to expand its team ๐ฅ and roll out its technology in partner data centers. With this growth, it is considering going public ๐ as it sees a huge opportunity in AI, and we're keeping our fingers crossed for them ๐ค!ย
ย
#4
AI in schoolwork ๐ค: Cheating case that ended up in court โ๏ธ๐.
In Massachusetts ๐บ๐ธ, the parents of a student are suing their son's high school ๐ซ after he was accused of cheating because he used artificial intelligence ๐ค when writing a history paper.ย
They claim the school had no clear rules ๐ for using AI and their son only used it to gather information, not to write the paper. However, the school handed out punishments including detention and lower grades ๐ which the family says may negatively affect the student's college applications ๐ and chances of getting into the National Honor Society ๐ (a student organization).
This case is one of the first โ๏ธ where the use of AI in schoolwork has led to legal action. It raises important questions โ about how artificial intelligence should be addressed in education ๐ฉโ๐ซ and whether schools are prepared to deal with the complex issues associated with this technology ๐. Schools and students need clearer rules as AI tools become more common. What do you think about this ๐ค?ย
ย
#5
Apple researchers reveal AI's weaknesses ๐ : Artificial intelligence relies on patterns, not logic, to solve problems ๐ค๐ง .
New research by Apple engineers reveals that the reasoning abilities of AI models are not as reliable as reported ๐ค๐ง .ย
The study found that even small changes in simple math tasks, such as swapping names or changing numbers, caused a significant drop in the performance of many AI systems ๐. With these seemingly trivial adjustments, the accuracy of high-end models such as GPT-4 ๐ฌ dropped. This suggests that AI relies heavily on pattern matching rather than truly understanding the logic of problems ๐ค.
The researchers also added irrelevant details to the questions, causing some models to make huge errors - accuracy dropped by a staggering 65% ๐ฑ. These findings highlight a major flaw in AI reasoning: instead of actually solving problems, AI tends to mimic solutions it has already seen, making its logic fragile and error-prone when faced with unknown scenarios ๐งฉ๐ฅ.
This calls into question the reliability of current AI models in tasks requiring real reasoning. What do you think about this ๐ง?
ย
