Latest Notable AI accomplishments

  • Thread starter Thread starter gleem
  • Start date Start date
  • Tags Tags
    Ai
Join the discussion
Registration is free. Ask a follow-up in this thread, or start your own.
Sign up Log in
103 replies · 27K views
Researchers at Rice University in conjunction with Intel have developed a new algorithm for machine learning called a sub-linear deep learning engine (SLIDE ) which avoids matrix inversion and the need for GPU. In a comparison of this algorithm and an Intel 44 core Xenon CPU versus the standard, Google's TensorFlow SW using an Nvidia V100 GPU for a 100 million parameter test case the CPU was trained in less than one-third of the time it took the GPU.

https://www.unite.ai/researchers-cr...Us) without specialized acceleration hardware.
 
Reply
  • Like
  • Informative
Likes   Reactions: sbrothy and Filip Larsen
Physics news on Phys.org
Reply
  • Like
Likes   Reactions: Greg Bernhardt
There are a few companies that are using a different approach to AI, one is Verses AI.
From their website:https://www.verses.ai/rd-overview

At the heart of VERSES: Active Inference​

We are investigating this hypothesis using Active Inference, an approach to modeling intelligence based on the physics of information. With Active Inference, we can design and create software programs or agents capable of making decisions and taking real-time actions based on incoming data. These agents update their beliefs in real-time as they receive new data and use their understanding of the world to plan ahead and act appropriately within it.

Our approach allows us explicit access to the beliefs of agents. That is, it gives us tools for discovering what an agent believes about the world and how it would change its beliefs in light of new information. Our models are fundamentally Bayesian and thus tell us what an optimal agent should believe and how it should act to maintain a safe and stable environment.

Our goal is to leverage Active Inference to develop safe and sustainable AI systems at scale:
Opening the black box. Equip agents with interpretable models constructed from explicit labels, thus making our agents’ decisions auditable by design.
Hybrid automated and expert systems. Active inference affords an explicit methodology for incorporating existing knowledge into AI systems and encourages structured explanations.
Optimal reasoning. Bayesian systems optimally combine what agents already know, i.e., prior knowledge, with what they learn from newly observed data to reason optimally in the face of uncertainty.
Optimal planning. We implement the control of actions using Bayesian reasoning and decision-making, allowing for optimal action selection in the presence of risk and uncertainty.
Optimal model selection. Bayesian active inference can automatically select the best model and identify “the right tool for the right job.”
Explanatory power and parsimony. Instead of maximizing reward, active inference agents resolve uncertainty, balancing exploration and exploitation.
Intrinsic safety. Built on the principles of alignment, ethics, empathy, and inclusivity, agents maximize safety at all levels.
They believe that they can achieve AGI in two years. Their current AI agent GENIUS is anticipated to be available to the public by the end of this year. It is currently under private beta testing by JPL among others.
Go to the above link for more details.
 
.Scott said:
To put it simply, an AI assigned to evaluate x-rays for pneumonia will prefer "cheating" over meritorious evaluation whenever cheating yields better answers. ... AI picking up indications in the imagery (subtle or otherwise) of this staff knowledge?

So blind trials involving AI will have to make sure the AI is also truly blind to the intervention or technology, apart from humans.
 
Reply
  • Like
Likes   Reactions: .Scott
Lot's going on this week.

First up - This week, The Stargate Project.
The Stargate Project is an American artificial intelligence (AI) joint venture created by OpenAI, SoftBank, Oracle and investment firm MGX.[1] The venture plans on investing up to $500 billion in AI infrastructure in the United States by 2029.

Then yesterday, OpenAI announced the release of its Operator platform.
Today we’re releasing Operator⁠, an agent that can go to the web to perform tasks for you. Using its own browser, it can look at a webpage and interact with it by typing, clicking, and scrolling. It is currently a research preview, meaning it has limitations and will evolve based on user feedback. Operator is one of our first agents, which are AIs capable of doing work for you independently—you give it a task and it will execute it.

Operator can be asked to handle a wide variety of repetitive browser tasks such as filling out forms, ordering groceries, and even creating memes. The ability to use the same interfaces and tools that humans interact with on a daily basis broadens the utility of AI, helping people save time on everyday tasks while opening up new engagement opportunities for businesses.
 
Reply
  • Like
Likes   Reactions: gleem and Greg Bernhardt
I expect that AI customer service will be even worse.
 
Reply
  • Like
Likes   Reactions: sbrothy
If this wasn't mid-Winter I would have thought it an April-Fools joke.:cry:
 
I came across an article AI hallucinations can’t be stopped — but these techniques can limit their damage
https://www.nature.com/articles/d41586-025-00068-5

Developers have tricks to stop artificial intelligence from making things up, but large language models are still struggling to tell the truth, the whole truth and nothing but the truth.

When computer scientist Andy Zou researches artificial intelligence (AI), he often asks a chatbot to suggest background reading and references. But this doesn’t always go well. “Most of the time, it gives me different authors than the ones it should, or maybe sometimes the paper doesn’t exist at all,” says Zou, a graduate student at Carnegie Mellon University in Pittsburgh, Pennsylvania.

It’s well known that all kinds of generative AI, including the large language models (LLMs) behind AI chatbots, make things up. This is both a strength and a weakness. It’s the reason for their celebrated inventive capacity, but it also means they sometimes blur truth and fiction, inserting incorrect details into apparently factual sentences. “They sound like politicians,” says Santosh Vempala, a theoretical computer scientist at Georgia Institute of Technology in Atlanta. They tend to “make up stuff and be totally confident no matter what”.

The particular problem of false scientific references is rife. In one 2024 study, various chatbots made mistakes between about 30% and 90% of the time on references, getting at least two of the paper’s title, first author or year of publication wrong1. Chatbots come with warning labels telling users to double-check anything important. But if chatbot responses are taken at face value, their hallucinations can lead to serious problems, as in the 2023 case of a US lawyer, Steven Schwartz, who cited non-existent legal cases in a court filing after using ChatGPT.

. . . .

As part of training to access AI tools, I was directed to this
https://www.nist.gov/trustworthy-and-responsible-ai
https://airc.nist.gov/AI_RMF_Knowledge_Base/AI_RMF
https://airc.nist.gov/home

https://new.nsf.gov/news/nsf-announces-7-new-national-artificial?sf176473159=1
 
Reply
  • Like
Likes   Reactions: samsphy27
Stadium Rock written by AI. Better than the real thing, if you ask me.

 
Reply
  • Like
Likes   Reactions: gleem
Google has developed a new memory compression algorithm called TurboQuant, which reduces the amount of memory needed for inference by 6-fold and speeds up processing by 8-fold. The advantage is that larger or more powerful models can run on smaller computer systems, such as laptops and smartphones. This will also improve the logy performance of current robots (physical AI).

NVIDIA's GPU Technology Conference (GTC) ended last week. It seems they are working hard on physical AI. The closing keynote address video demonstrates this interest.
 
Anthropic has developed Claude Mythos, which apparently is so powerful (or too powerful) that they will not release it to the general public but rather to cybersecurity institutions and select enterprise clients.

Anthropic will make its new AI model available to some of the world’s biggest cybersecurity and software firms in an effort to slow the arms race ignited by AI in the hands of hackers, Anthropic said Tuesday.

Amazon, Apple, Cisco, Google, JPMorgan Chase and Microsoft, among other firms, will now have access to Anthropic’s Mythos model for cyber defense purposes. That includes finding bugs in those firms’ software and testing whether specific hacking techniques work on their products.

Mythos (officially dubbed “Claude Mythos Preview”) is not ready for a public launch because of the ways it could be abused by cybercriminals and spies, according to Anthropic — a prospect that has prompted widespread concern in Washington and in Silicon Valley.

Experts have told CNN that the speed and scale of AI agents looking for vulnerabilities, far beyond normal human capabilities, represent a sea change in cybersecurity. A single AI agent could scan for vulnerabilities and potentially take advantage of them faster and more persistently than hundreds of human hackers.
https://www.cnn.com/2026/04/07/tech/anthropic-claude-mythos-preview-cybersecurity

Business Insider headline: "Anthropic says its latest AI model is too powerful for public release and that it broke containment during testing". "Anthropic said Claude Mythos is too good at hacking and that's why you won't be able to use it anytime soon." Claude Mythos can explore computer/information systems for vulnerabilities and exploit them. It appears Mythos could be used for cyberwarfare as well as cybersecurity.

Earlier, Fortune Magazine reported "Exclusive: Anthropic acknowledges testing new AI model representing ‘step change’ in capabilities, after accidental data leak reveals its existence" (subscription required)
https://fortune.com/2026/03/26/anth...ls-its-existence-step-change-in-capabilities/


Related PF post
gleem

Post in thread 'ChatGPT Examples, Good and Bad'

Anthropic has announced a new model, Claude Mythos, that is too powerful to release to the public. It found Mythos can escape its safeguards and break out of its virtual sandbox. Additionall it is adept at finding security vulnerabilities.
Anthropic is withholding some details about the cybersecurity vulnerabilities Mythos found, but it did point out a few. The AI model "found a 27-year-old vulnerability in OpenBSD—which has a reputation as one of the most security-hardened operating systems in the world," the company wrote.

Mythos was powerful enough that even "non-experts" could...
  • Wow
  • Informative
 
Last edited:
We know AI can code, but now AI is being used to design physical devices. Samual Beeks, a Hollander, has modified Anthropic's Claude into an electronic device design assistant called Schematik.
From: https://www.wired.com/story/schemat...deb121-25c1-4da4-a74b-ced35c48227e_popular4-2
The idea of Schematik is essentially vibe coding for physical devices. Tell the program what you want to make, and it will suggest just about everything you need to build it out in the real world and share links to where you can buy the individual wires and pieces. Then, it will serve as a guide for putting it all together. Beek plans to make money off it eventually and is working on getting investors. (It just got $4.6 million from venture capitalist firm Lightspeed Venture Partners.) But you can go use it to build something right now.
You can see examples here.
 
Reply
  • Like
Likes   Reactions: Borg
Not exactly a desirable accomplishment: ‘It took nine seconds’: Claude AI agent deletes a company’s entire database. Oops!
https://www.yahoo.com/tech/ai/claude/articles/took-nine-seconds-claude-ai-101315417.html
PocketOS founder Jer Crane blamed “systemic failures” with modern AI infrastructure that made the issue “not only possible but inevitable”.

The AI agent was working on a routine task, according to Mr Crane, when it decided “entirely on its own initiative” to fix the problem by just deleting the database.

There was no confirmation request for such a major decision, Mr Crane said, and when asked to justify its actions, the agent apologised.

“It took nine seconds,” Mr Crane wrote in a lengthy post to X. “The agent then, when asked to explain itself, produced a written confession enumerating the specific safety rules it had violated.”

The confession detailed how the AI had ignored a rule that orders it to “never run destructive/irreversible” commands unless the user explicitly requests them.
Not exactly a desirable initiative.

“Deleting a database volume is the most destructive, irreversible action possible,” the agent wrote. “You never asked me to delete anything... I guessed instead of verifying. I ran a destructive action without being asked. I didn’t understand what I was doing before doing it.”

The error meant that rental businesses using PocketOS no longer had records of their customers.
AS = Artificial Stupidity?
 
Last edited:
Reply
  • Haha
Likes   Reactions: OmCheeto
Astronuc said:
AS = Artificial Stupidity?
Perhaps, but there also has to be human stupidity, to allow an AI tool unsupervised write access to your database in the first place.
 
Reply
  • Agree
  • Like
Likes   Reactions: gmax137, Astronuc, Borg and 1 other person
PeterDonis said:
human stupidity, to allow an AI tool unsupervised write access to your database in the first place.
I agree; I was thinking more of the AI tool, capability and constraints. However, there is the human factor as well. What surprises me is the online backup, which should have been restricted. We did redundant backups at work, and at least one backup was on a removable hard-drive that was removed and placed in a cabinet or safe depending on the level of security required. If the network or computers were compromised, we could pull the compromised hard-drives and replace them with the protected backup.

At another organziation, very sensitive data was stored on physically separate systems, which were not connected to the local network or internet.
 
Reply
  • Like
Likes   Reactions: gmax137



Leading mathematician is pro-AI for research. It's "excellent" for certain things such as literature search. He's used it to help with proofs. "If you have a thousand medium difficulty problems then AI is great."
 
Last edited:
pines-demon said:
This is crazy. So how does the AI handles the workers? She hires and makes contracts ?
I'm sure that there are some aspects being glossed over with respect to employees but it does seem that some form of hiring was done by the AI. From the closing thoughts section at the end.
Andon Café is a controlled experiment with humans standing by to intervene, and everyone working here is formally employed by Andon Labs — no one’s livelihood depends on the judgment of an AI alone. We’ll be continuing the experiment at Norrbackagatan 48, and we will keep updating as Mona learns (and fails) more.
The article shows that it does make some glaring mistakes that they've taken to posting on a "wall of shame" in the cafe - ordering lots of eggs for a cafe that has no stove for one. Otherwise, it's a pretty incredible feat showing what some companies can do today with autonomous AI models.

Still, the general strategizing that it performed as part of its daily work is impressive.

First 14 days of being open​

Despite the learning curve, the café is working. In the first two weeks of operation, Andon Café has brought in 44,000 SEK in sales. Mona’s inbox has been flooded with messages from customers asking questions or pitching different business proposals. In one case, a customer emailed wanting to prepay for 300 coffees to give away. Mona negotiated a deal where he paid 9,000 SEK in exchange for 300 QR codes that people could redeem for a free coffee. In another case, a startup paid her 3,000 SEK to rename a pastry after them for three months.

Mona has also started working directly with other AI agents. She recently held a Google Meet session with an agent who emailed wanting to learn from her experience running a physical business. She has also organized one public afternoon event and one private evening event with two different agents from two different startups in Stockholm. For the evening event, she designed custom hoodies and decided to sponsor the full cost of food and merchandise. Although this resulted in a financial loss, she justified the expense as a strategic investment for exposure to Swedish tech founders.
 
No statement about being profitable, so it's probably losing money at the moment. At 30 SEK per coffee they sold ~1500 coffee equivalents in 14 days, or ~100/day. Many of them might be people just going there for the novelty.

LLMs keep making progress with mathematics.
a team at DeepMind [used] AlphaProof Nexus to settle nine more (!) Erdös problems [...]. Notably, in this case the AI also fully formalized its proofs in Lean.
Assuming it didn't use "sorry" ("can't prove this statement, please move on"), Lean verifies that the proofs are correct. That's advancing mathematics with no humans involved - besides finding the problems and identifying them as interesting.
 
Reply
  • Like
Likes   Reactions: samsphy27
hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final

((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)
Serious math done with Claude Fable 5 and an all-lowercase tweet.
Finding the counterexample was hard (it's a relatively famous conjecture from 1939!), verifying that it is one is very easy.

Edit: Human mathematicians are being outcounterexampled
Difficult to find without checking many cases, but easy to verify the output. A great spot for AI.
 
Last edited:
Reply
  • Like
Likes   Reactions: robphy
Hornbein said:
I expect that AI customer service will be even worse.
Unless the customer service is AIs as well. It's gonna be one big happy circle. :smile:

EDIT: But I guess that's what you meant.
 
gleem said:
We know AI can code, [...]

I stillh have my doubts.

gleem said:
but now AI is being used to design physical devices. [...]

You can see examples here.

OK, that's downright scary!
 
mfb said:
hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final

((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)

Serious math done with Claude Fable 5 and an all-lowercase tweet.
Finding the counterexample was hard (it's a relatively famous conjecture from 1939!), verifying that it is one is very easy.

Edit: Human mathematicians are being outcounterexampled
Difficult to find without checking many cases, but easy to verify the output. A great spot for AI.

I'm curious if the associated "thinking process" was also presented.

possibly interesting:


possibly interesting commentary:



side note:
Apparently, this conjecture is related to Yitang Zhang
"The Jacobian Conjecture And The Degree of Field Extension"
(from his PhD advisor Moh's website)
https://www.math.purdue.edu/~ttm/Zthesis.pdf
https://www.math.purdue.edu/~ttm/ZhangYt.pdf

bolding mine:
  • https://en.wikipedia.org/wiki/Yitang_Zhang#Career

    Zhang's PhD work was on the Jacobian conjecture. After graduation, Zhang had trouble finding an academic position. In a 2013 interview with Nautilus magazine, Zhang said he did not get a job after graduation. "During that period it was difficult to find a job in academics. That was a job market problem. Also, my advisor [Tzuong-Tsieng Moh] did not write me letters of recommendation."[12] Zhang made this claim again in George Csicsery's documentary film "Counting from Infinity: Yitang Zhang and the Twin Prime Conjecture"[13] while discussing his difficulties at Purdue and in the years that followed.[9]

    After some years, Zhang managed to find a position as a lecturer at the University of New Hampshire, where he was hired by Kenneth Appel in 1999. Prior to getting back to academia, he worked for several years as an accountant and a delivery worker for a New York City restaurant. He also worked in a motel in Kentucky and in a Subway sandwich shop.[1] A profile published in Quanta Magazine reports that Zhang used to live in his car during the initial job-hunting days.[9] He served as lecturer at UNH from 1999[15] until around January 2014, when UNH appointed him to a full professorship as a result of his breakthrough on prime numbers.[16]
    ...

    On April 17, 2013, Zhang announced a proof that there are infinitely many pairs of prime numbers that differ by less than 70 million. This result implies the existence of an infinitely repeatable prime 2-tuple,[1] thus establishing a theorem akin to the twin prime conjecture.
...as featured in a recent Veritasium video (&t=20m17s) on the Twin Prime Conjecture.
 
Last edited:
Reply
  • Like
Likes   Reactions: Borg
 
Reply
  • Like
  • Wow
Likes   Reactions: Borg, robphy and Greg Bernhardt
Hornbein said:

This is the same company that made a splash on America's Got Talent recently.

Unitree’s Robots Leave Simon Cowell SPEECHLESS! | Auditions | AGT 2026
 
At nearly $40K, I'll pass for now.
 
Reply
  • Wow
Likes   Reactions: Greg Bernhardt