Google says 25% of its code is AI-generated

Yesterday the world passed some sort of watershed moment during Google’s quarterly earnings call with financial analysts. CEO Sundar Pichai casually mentioned the following startling statistic: over 25% of all new computer code implemented at Google is now produced by AI. Human coders have to approve the code before it is put into production, but… still. Google is the inventor of the modern Transformer AI architecture, and a major producer of AI models and services. So now we have AI writing the code to improve AI so that AI can write better code to improve itself more - and on and on. Like it (I mostly do) or not, we may well see exponential improvements in AI capabilities over the next decade or more.

Google CEO says 25% of its new computer code is being generated by AI

Microsoft warns of foreign disinformation hitting US election

Last week Microsoft’s Threat Analysis Center (MTAC) warned that malicious foreign influence operations based in Russia, China, and Iran continue to be very active, and are continually evolving in an attempt to increase their success in shaping the results of the current US presidential election. Iran is mainly targeting the Trump campaign, and trying to promote anti-Israel sentiment in the US. Russia, on the other hand, is more focused on the Harris campaign, spreading character attacks and scurrilous deepfakes. China has been active in down-ballot races, targeting candidates for the House or Senate who have espoused positions unfavorable to China’s interests. Microsoft emphasizes that it is all but certain that these malign actors will try to sow distrust in the legitimacy of the results of the election, since division among the American people weakens the US abroad.

Microsoft warns of foreign disinformation hitting US election from all sides

TSMC’s chip plant in Arizona has better yields than those in Taiwan

The most advanced computer chips in the world are made by a single company, TSMC of Taiwan, which is perched perilously near the coast of China. Biden’s CHIPS Act has put billions into trying to repatriate chip production to the US, on both national security and economic grounds, and now we have some of the first fruits of that program. TSMC is building 3 large advanced factories in Arizona, and the first one is now producing chips. A quality check now shows that not only are the US chips good - they have fewer defects than chips from the parent factory in Taiwan. This is a stunning achievement for all involved, and makes the world a slightly safer place.

TSMC chip plant in Arizona has better yields than the original in Taiwan

Apple Intelligence debuts - and fails to wow

On Monday, Apple rolled out its first round of operating system upgrades that will enable its products to use AI, and the result was - nice, but a bit underwhelming. The most anticipated AI was a chattier, more conversational Siri, but she still can’t order a pizza for you. Compared to the flirty agent of OpenAI’s Advanced Voice feature, or the crisper, more buttoned-down assistant of Google’s free Gemini Live app, Siri seems a bit of a letdown. Apple is rolling out a number of non-Siri AI features in its apps, which will no doubt be useful, but nothing that is really better than or even different from its rivals. CEO Tim Cook vows that Apple will eventually be “not the first, but the best” in on-device AI, and he has over a billion users worldwide that fervently hope he is right about that.

Apple Intelligence isn’t ready to wow you - yet

14-year-old wins 3M science challenge with AI pesticide detector

14-year-old Sirish Subash has been named America’s Top Young Scientist in the annual 3M Young Scientist Challenge. The teen’s prize-winning project was the development of a handheld device that uses AI to analyze the light waves bouncing off fresh produce to identify the type and amount of any pesticide residue on the surface. Such residues can have negative effects on the consumer’s health, and easy identification of such contaminants can help protect the public from these harms.

14-year-old wins 3M science challenge with AI-powered pesticide detector

Biden issues directive for federal agencies to accelerate use of AI

Providing more evidence that the Biden White House considers AI to be a national security issue, President Biden issued a directive last week that all federal agencies must accelerate use of AI, while developing the necessary safeguards to avoid bias or harm. The directive was framed as a means for assuring that the US remains preeminent in AI in the world, which some in the administration view as an existential necessity for world peace and US prosperity.

Biden issues directive for federal agencies to accelerate use of AI

Universal partners with Klay to produce “ethical AI music”

Universal Music Group has signed an agreement with AI music startup Klay Vision. The partnership aims not to replace human musicians with AI, but to develop “ethical AI” which allows creators to use AI in their creations. This comes against a backdrop of a groundswell of opposition to AI among singers, songwriters, musicians, and creative talent from other fields.

Universal partners with Klay to produce “ethical AI music”

AI performs literature reviews better than PhDs

PaperQA2 is an open-source AI project that assists researchers to perform literature searches on a given topic. Given a topic, the AI will independently search for relevant articles, categorize them, identify conflicts, and produce a summary of the relevant findings. A recent study compared the accuracy and completeness of the AI’s results to those of human PhDs - and the AI beat the humans handily, achieving “superhuman synthesis of scientific knowledge” according to the oh-so-modest researchers.

AI performs literature searches better than PhDs

arXiv scientific paper

AI transcription tool used by thousands of MDs makes stuff up

“Ambient documentation”, using AI to produce a transcript of a physician-patient interaction and then to summarize it for the patient record, is sweeping the healthcare field in the US, to the point where it is starting to become the standard of care in certain large health systems. Now a new series of studies is finding that one basic building block of some such systems, an open-source AI-powered transcription system called Whisper, can actually produce text in the transcript that is completely absent in the audio record. In other words, Whisper sometimes just makes stuff up. In the AI field, this is known as “the hallucination problem”, and all of the major AI companies have spent literally billions of dollars to try to eliminate it in their products. Whisper, on the other hand, is open-sourced (which means free to use) and so has been adopted by some smaller companies as a quick and cost-effective way to get a product to market. Now there are tens of thousands of physicians using these Whisper-based products every day to document patient interactions, raising the concern that inaccurate results may be placed in the medical record. Each physician is legally responsible for editing and verifying the AI’s output, but there is understandable concern that AI errors might slip through and cause harm to patients. A shakeout of the AI medical documentation sector may be coming, in which only provably accurate systems may survive, and this may be no bad thing.

AI transcription tool used by thousands of physicians “hallucinates”

AI-designed “DNA switch” flips genes on or off

Researchers at MIT, Harvard, and Yale have used AI to design thousands of DNA switches that can precisely control the expression of a given gene in different cell types. This will allow clinicians and researchers to control the expression of a gene in just one tissue, without affecting the rest of the body. This will make possible more precise experiments, as well as more finely targeted therapies for medical conditions.

AI-designed “DNA switch” flips genes on or off

Researchers target childhood heart disease with AI and stem cells

Murdoch Childrens Research Institute of Australia has forged a collaboration with the Gladstone Institute of San Francisco to understand, prevent, and cure childhood heart disease using stem cell technology and AI. The aim is to prevent progression of the disease and avoid the need for a heart transplant.

Researchers target childhood heart disease with AI and stem cells

AI predicts antimicrobial resistance in real time from the EHR

Antimicrobial resistance is a pervasive problem in clinical settings, particularly in hospitals and ICUs, where heavy antibiotic use creates pressure for microbes to evolve resistance. Researchers in Abu Dhabi, UAE, have devised an AI system that continuously monitors the EHR and predicts possible emerging resistance. This appears to be a cost-effective means for predicting resistance in real time, as opposed to waiting to identify resistance after it has emerged.

AI predicts antimicrobial resistance in real time from the EHR