Tech giants were cautious with advanced facial recognition - AI startups are not

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each weekday, we scour the web to aggregate the many stories related to artificial intelligence; we pass along the news, useful resources, tools or services, technical analysis and exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on under the hood in AI.

(Opinion) AI and the end of programming?

The Story: Matt Walsh, a former Professor of Computer Science at Harvard and Google/Apple team leader, has stated unequivocally that the end of classical computer science is coming and that programming will be obsolete. He theorizes a window between 10-30 years, after which there will be little need for human involvement. The headline opinion piece above provides an analysis of the current landscape through the lens of the current capabilities of ChatGPT (3.5 and 4)

More Details:

• Hayes asked ChatGPT to solve word ladder puzzles, where you change one letter at a time to transform one word into another. ChatGPT made errors like repeating words and allowing invalid words. GPT uses brute-force algorithms, even when a more elegant solution is easily available.

• When Hayes prompted ChatGPT to write code to solve problems like calculating Fibonacci numbers, it produced wrong solutions or code that simply fails.

• Studies by Hayes and others found AI-generated code failed 30-50% of the time on basic programming tasks. Performance was worse on more recent problems missing from the AI's training data. The code lacked reliability and robustness.

• Hayes argues the statistical, language-modeling nature of systems like ChatGPT lacks true reasoning ability. Until an AI system has that, humans will be needed as much as ever.

Takeaways: Current AI systems may appear impressive (and they are!) but simply cannot yet match human coding skills in creative problem solving and debugging. Rote code generation divorced from meaning has very clear limitations in the real world.

But that’s only the present moment. Looking ahead, strong AI or artificial general intelligence (AGI) is the point at which an AI could learn to effectively accomplish any intellectual task that human beings can perform. We are not yet at AGI; no one knows when will get there, but most experts are certain that it’s coming.

Superintelligence implies an even more extreme form of intelligence that we cannot fully comprehend or control, beyond AGI. AGI does not necessarily have to reach this profound level to be considered capable of replacing humans. But OpenAI believes even superintelligence could arrive this decade. Paying attention to the trends, and planning around them, will be critical to anyone who doesn’t wish to be left behind - that goes for more than just programming.

In the case of PimEyes, the only way to get yourself out of their system is to first even know about the service in the first place, and then go through a user-submission process to explain why you don’t want to be in their database. Here’s the real kicker, direct from the website:

“Note: Review of the request and complete removal of the search result takes about 48 hours. So if your request doesn't get approved, the search image will reappear in search results.”

Why can they deny your request to publicize your data?

• PimEyes operates as a Polish company outside US jurisdiction. Even if they violated American privacy law - enforcement would be very difficult.

• Were they US-based, specific federal laws regulating facial recognition technology and its uses are still currently lacking, although some states have legislation pending. Without clear restrictions, companies can gather and monetize biometric data. That might not matter as much when it’s a FitBit, but the stakes are increasing.

• Datasets and algorithms shared freely for ‘academic purposes’ - and open data scraped from the web - can currently be repurposed for commercial applications like PimEyes without consent.

Students will take a 1 year leave from medical school to complete AI coursework at UTSA. The degree combines medicine curriculum with courses in data analytics, computer science, and intelligent systems. They aim to prepare graduates to lead advances in research, education, industry and health care administration.

While it’s not a combined degree, Harvard also recently announced a PhD track for Artificial Intelligence in Medicine (AIM).

• Majority of US employees are using AI at work, survey finds

• Tech leaders agree on AI regulation but divided on how in Washington forum

• ‘Thursday Night Football’ will use AI to help viewers better understand the action on the field

• US defense chief insists world 'nowhere close' to existential AI threat

They see AI improving quality and speeding up production but not significantly reducing costs or alleviating talent shortages. Executives highlighted the continued importance of human creativity, viewing AI as an enhancing tool requiring oversight.

They expect a larger impact on pre-production and production vs. other stages. Challenges include system integration, data training, capabilities, legal issues, and costs. Bain recommends a disciplined AI approach focused on benefiting players, strategically building capabilities, and adapting processes.

• (Open source) Instagraph: go-to application for converting text or URLs into insightful knowledge graphs

• (Open source) LangChain Teacher

• (Open source) HuggingFace announces a more efficient text-to-image generator than Stable Diffusion

• Introducing Replit ModelFarm

• Product-led AI - backing the gold miners