Sam Altman says Open AI knows how to build AGI, now setting sights on ASI

AGI, or Artificial General Intelligence, refers to machine intelligence that equals that of an average human being. It has been a fond hope of many an AI researcher for decades, and it was the major driving vision behind OpenAI’s founding, as well as its development of ChatGPT (as just as a step along the road, you see.) With the announcement of its latest “reasoning” model, named o3, which actually passed one well-known test for human intelligence (see story below), CEO Sam Altman is declaring victory. In a blog post, he writes that OpenAI now knows how to build a model with AGI, and sooner or later, they will. He now is setting his sights on the next step up the intelligence curve, ASI - Artificial Super-Intelligence, a machine that is smarter than the smartest human. Hey, a guy’s gotta dream, right?  And Sam always dreams big.

Sam Altman’s post as reported in The Verge

Sam Altman blog saying OpenAI knows how to build AGI, sets sights on ASI

Nvidia CEO Jensen Huang wows the crowd at CES

Jensen Huang, legendary CEO of AI chip powerhouse Nvidia, gave a crowd-pleasing keynote address at this week’s massive Consumer Electronics Show in Las Vegas. The talk was full of juicy reveals, but here are two of the main takeaways:

1. A supercomputer on your desk: Nvidia will begin producing desktop computers in mid-2025 that can run AI models with 200 billion parameters - this is larger than all but the largest AI models from OpenAI, Google, Meta, and Anthropic. The processing power will be a jaw-dropping 1 Petaflop (a million billion floating point operations per second.) For comparison, Microsoft and Apple’s top AI personal computers will sport 40 Teraflops (40 thousand billion operations per second, 1/25th of the power of the Nvidia machines, which are projected to cost a pricy but not crazy $3,000.

2. Robots and agents are a trillion-dollar future. Nvidia is heavily supporting processing chips and software for robots, which for Huang include humanoid factory worker bots and self-driving cars. Huang also promoted the idea that semi-autonomous “agents” that can actually perform tasks for you with minimal guidance are the Next Big Thing.

Jensen Huang CES keynote takeaways

OpenAI’s reasoning model aces human intelligence test, but judges move the goalposts

OpenAI produced a flurry of upgrades and new models in December, and for a finale they announced an advanced reasoning model, known as o3. OpenAI’s most recent models have struck out in a new direction, increasing performance not by training on more data, but by engineering the model to consider questions step by step and to check its work. o3 does this more and better than any AI model before it, and this has caused such an improvement in performance that it managed to score 87% on a test specifically designed to detect human-level intelligence, the ARC-AGI test. (AGI stands for Artificial General Intelligence, the term used for human level intelligence produced by machines.) Actual humans only score an average of 85% on this test. So hooray, hooray, OpenAI wins the $1,000,000 ARC Prize and su-u-per bragging rights, yes? Well, no. The ARC Prize Committee freaks out, says “No fair, you cheated!” rewrites the rules, and is revising the test. Sore losers. Anyway, reasoning models seem destined to be a big part of AI in the future, because the new method does seem to produce impressive results, especially in math and computer coding.

OpenAI’s reasoning model aces human intelligence test, but gets no prize

Meta drops content moderation, deletes bots

Elections have consequences. Meta (Facebook) has just shut down its content moderation team, the folks who work to make sure that hateful or misleading content does not get amplified on Meta’s social media sites such as Facebook and Instagram. Incoming President Trump has been relentlessly vocal against such moderation, seeing it as censorship of conservative opinions. Meta founder-CEO Zuckerberg has now thrown in the towel, and Meta’s social media sites may become free-fire zones for right wing “debate.”

Yet public opinion still matters. Just days after announcing that Meta sites will allow creation of AI-generated fake people to hold accounts and post content, Zuck reversed himself and shut down all of the handful of such “personalities” that were already existing. The reason? The immediate and near-universal revulsion expressed by users of the platforms at this truly awful idea.

Meta drops content moderation

Zuck deletes the bots due to public outcry

Chinese DeepSeek AI model impresses with power and efficiency

Chinese AI company DeepSeek has released an AI model that equals or exceeds the performance of all the top models from US giants OpenAI, Google, Meta/Facebook, and Anthropic. That’s impressive enough, but the company claims that they trained this phenomenon in only 2 months, using older, slower computer chips, and it cost only about $5 million dollars (a fraction of what US companies spent using the latest chips.) If true, this has several implications:

1. China refuses to be hobbled by US chip embargos: despite Biden’s vigorous efforts to cut off the flow of advanced AI chips to China, the country is finding creative ways to stay in the AI race.

2. We are still in the early days of developing AI technology, and creative types may find ever cheaper, faster ways of advancing the state of the art. This will blunt the advantage of the current AI incumbents, and allow new entrants to compete.

3. DeepSeek has made this model open source, meaning that anyone in the world can run it on their own hardware, if they have the knowhow and the (large but not massive) bankroll. Again, this helps even the playing field for smaller, newer entrants.

Chinese DeepSeek AI model impresses with power and efficiency

Samsung demos AI fridge at CES Las Vegas

South Korean electronics giant Samsung is showing off a fleet of AI-infused products at this week’s massive Consumer Electronics Show in Las Vegas. Notably, they have developed an AI refrigerator, which can identify food on its shelves with internal cameras, then use that AI-produced inventory to suggest tasty recipes that use those ingredients, or start a shopping list of replacements needed as food is eaten. Samsung’s partnership with Instacart will even allow the AI fridge to order the needed items and have them delivered to your door. Feeling hungry? Talk to your refrigerator.

Samsung demos AI fridge at CES Las Vegas

AI artist “Botto” makes millions selling its images at auction

"Botto,” an AI “machine artist” designed and built by a collaboration between computer programmers and human artists, has sold its art at auction for a combined total of over $5 million since its beginning in 2021. Each week, Botto produces multiple images, and they are voted on by the 5000 members of the BottoDAO - or distributed autonomous organization. The winning image for the week is then put up for auction, and the proceeds are split up among the voting members after infrastructure costs are paid. Last October, two early images created by Botto fetched $276,000 at auction at Sotheby’s New York.

AI artist “Botto” makes millions selling its images at auction

Penn State AI predicts progression of autoimmune disease

Autoimmune diseases, where the immune system mistakenly attacks the body’s healthy tissues, are common, affecting some 8% of the US population. These diseases often have a preclinical phase, characterized by mild symptoms or just some unusual antibodies in the blood. Many more patients have these subclinical symptoms and blood tests than ever go on to have the full-blown condition. Determining which of these patients will go on to develop overt disease could allow physicians to concentrate resources on the most at-risk patients, to slow or halt disease progression. Now researchers at Penn State have developed an AI system that examines patient information in the medical record, plus results of genetic testing, to identify patients with a high risk of developing overt or even severe disease. When tested on real world data of patients with Rheumatoid Arthritis or Lupus, the model outperformed a field of 20 other models at predicting which patients were most likely to have their disease progress. Researchers are now working to turn these AI-enabled clinical insights into better care for patients at risk.

Penn State AI predicts progression of autoimmune disease

Epic’s 2025 AI priorities

Epic, the dominant electronic medical record system for large health systems in the US, is all in on AI. Their priorities for development in 2025 include:

• Implementing AI Scribes that produce summaries of physician-patient interactions in real time, which the physician can quickly edit into a chart note.

• Adding voice order entry, and clinical decision support. Once the AI is listening to the physician-patient interaction, it is possible to enable the physician to enter orders by voice. Going further, the AI could be a real-time clinical assistant, suggesting to the physician possible diagnoses, tests for workup, and referrals to specialists based on the current clinical situation.  

• Translating clinical notes and reports into terms the patient can understand. Federal law mandates that patients have access to their medical records. AI can translate from “doctor speak” into accessible language.

• Helping physicians and patients surmount the barriers to care put up by health plans, such as prior authorization procedures. AI can be a major help in cutting through red tape and jumping through bureaucratic hoops.

Epic’s 2025 AI Priorities

Canadian AI can detect high blood pressure from your voice

Researchers from the University of Toronto have developed an AI model that can diagnose high blood pressure from speech recordings. The team recruited subjects to record brief voice recordings multiple times per day over several days. They then developed an AI model that classified subjects as either hypertensive or normotensive based only on characteristics detected in the voice recordings. Preliminary results of a pilot study of 245 individuals found that the AI achieved an accuracy in diagnosis of approximately 80%. The same researchers are trying to extend these results to predict blood sugar levels in diabetic patients, based only on voice recordings. These results point to a day when it may be possible for calls with your doctor to include a virtual physical exam based only on your voice and video.

Canadian AI can detect high blood pressure from your voice

Tiny shape-shifting robot is being developed for microsurgery

Researchers at the University of Colorado Boulder are building tiny robots, inspired by insects, that can actually morph their shape to squeeze through tight passageways - like inside the human body. As can be seen in the picture below, a penny-sized robot with 4 independently controllable legs, can wiggle, push, and actually compress itself to progress to a predetermined target - like a tumor, or a clot. These robots are only prototypes, but the researchers are working to make the concept workable in clinical medicine.

Tiny shape-shifting robot being developed for microsurgery