Nobody cares about "AI" (Chatbot: I disagree.)

[Noodles]

Well, this is an interesting turn of events; I imagine it is safe to assume the CAPTCHA's AI will now go to work beating all the other AI's, leading to the inevitable escalation on all sides.

I guess we can all sit back with a refreshing beverage of our choice and watch these idiot machines render each other useless.

Now, when it asks to "click all the motorcycles " it will only know you are not a robot because you clicked all the busses instead.

 

[Dakota Tebaldi]

Google has introduced AI podcast bros and girls! Actually it's a free service called NotebookLM, where you can upload documents to or feed it some web site links, and based on that content it will automatically generate a radio style podcast between a male and female host using AI cloned voices, talking about some topics out of the uploaded content.

So without further ado I proudly present: the VVO podcast!

This is kind of funny - the AI clearly sees a few post topics and it mentions them because it has seen them, but its commentary on controversial topics is super shallow because it probably has been programmed to not say anything that could be construed as an opinion. So like, it'll get into detail about people's posts in the Masquerade Ball thread, but then it'll mention that the forum has talked about Elon Musk but won't say what we've said about him, lol.

 

[Free]

The lesson is: AI takes itself way too serious.

AI digests repetitive scatological document into profound “poop” podcast

What happens when you ask an AI model to summarize a bunch of crap?

arstechnica.com

Imagine you're a podcaster who regularly does quick 10- to 12-minute summary reviews of written works. Now imagine your producer gives you multiple pages of nothing but the words "poop" and "fart" repeated over and over again and asks you to have an episode about the document on their desk within the hour.

Speaking for myself, I'd have trouble even knowing where to start with such a task. But when Reddit user sorryaboutyourcats gave the same prompt to Google's NotebookLM AI model, the result was a surprisingly cogent and engaging AI-generated podcast that touches on the nature of art, the philosophy of attention, and the human desire to make meaning out of the inherently meaningless.

 

[Isabeau]

Apple backs out of backing OpenAI, report claims

Apple dropped out of the $6.5 billion investment round at the 11th hour.

arstechnica.com

 

[Free]

Surprise, surprise.

Devs gaining little (if anything) from AI coding assistants

Code analysis firm sees no major benefits from AI dev tool when measuring key programming metrics, though others report incremental gains from coding copilots with emphasis on code review.

www.cio.com

Coding assistants have been an obvious early use case in the generative AI gold rush, but promised productivity improvements are falling short of the mark — if they exist at all.

Many developers say AI coding assistants make them more productive, but a recent study set forth to measure their output and found no significant gains. Use of GitHub Copilot also introduced 41% more bugs, according to the study from Uplevel, a company providing insights from coding and collaboration data.

The study measured pull request (PR) cycle time, or the time to merge code into a repository, and PR throughput, the number of pull requests merged. It found no significant improvements for developers using Copilot.

 

[Bartholomew Gallacher]

Nvidia has released its own first LLM with a big bang: it's totally open source and even comes with all the training data included! Performance wise it is rivaling GPT4.

So tough times for OpenAI, but for Nvidia it doesn't matter to give it away for free because as long as all are running their stuff on Nvidia hardware they can afford to give this out for free.

Nvidia just dropped a bombshell: Its new AI model is open, massive, and ready to rival GPT-4

Nvidia has released NVLM 1.0, a powerful open-source AI model that rivals GPT-4 and Google’s systems, marking a major breakthrough in multimodal language models for vision and text tasks.

venturebeat.com

 

[Noodles]

Surprise, surprise.

Devs gaining little (if anything) from AI coding assistants

Code analysis firm sees no major benefits from AI dev tool when measuring key programming metrics, though others report incremental gains from coding copilots with emphasis on code review.

www.cio.com

Personally, it feels like at best its going to result in exteemely insecure code. Because 90% of what its probably igested is tutorials, which almost never discuss security concerns and methods around coding.

 

[Khamon]

Surprise, surprise.

Devs gaining little (if anything) from AI coding assistants

Code analysis firm sees no major benefits from AI dev tool when measuring key programming metrics, though others report incremental gains from coding copilots with emphasis on code review.

www.cio.com

What Noodles said. The AI might produce a plausible solution sooner than the human, or vice-versa, but the examination and testing phases are going to take the same amount of time. And the hoomun will work at a comfortably consistent pace using the tools available. Yes. We Scottyize always.

 

[Free]

Sophisticated AI models are more likely to lie

Human feedback to AIs makes them favor providing an answer, even a wrong one, while making the answer more convincing.

arstechnica.com

“The notorious problem with reinforcement learning is that an AI optimizes to maximize reward, but not necessarily in a good way,” Schellaert says. Some of the reinforcement learning involved human supervisors who flagged answers they were not happy with. Since it’s hard for humans to be happy with “I don’t know” as an answer, one thing this training told the AIs was that saying “I don’t know” was a bad thing. So, the AIs mostly stopped doing that. But another, more important thing human supervisors flagged was incorrect answers. And that’s where things got a bit more complicated.

AI models are not really intelligent, not in a human sense of the word. They don’t know why something is rewarded and something else is flagged; all they are doing is optimizing their performance to maximize reward and minimize red flags. When incorrect answers were flagged, getting better at giving correct answers was one way to optimize things. The problem was getting better at hiding incompetence worked just as well. Human supervisors simply didn’t flag wrong answers that appeared good and coherent enough to them.

In other words, if a human didn’t know whether an answer was correct, they wouldn’t be able to penalize wrong but convincing-sounding answers.

"AI" companies are trying to build expert systems without any actual experts involved with their training.

 

[Free]

Some Apple researchers are dumping on the actual intelligence part of AI.

Apple study exposes deep cracks in LLMs’ “reasoning” capabilities

Irrelevant red herrings lead to "catastrophic" failure of logical inference.

arstechnica.com

For a while now, companies like OpenAI and Google have been touting advanced "reasoning" capabilities as the next big step in their latest artificial intelligence models. Now, though, a new study from six Apple engineers shows that the mathematical "reasoning" displayed by advanced large language models can be extremely brittle and unreliable in the face of seemingly trivial changes to common benchmark problems.

The fragility highlighted in these new results helps support previous research suggesting that LLMs use of probabilistic pattern matching is missing the formal understanding of underlying concepts needed for truly reliable mathematical reasoning capabilities. "Current LLMs are not capable of genuine logical reasoning," the researchers hypothesize based on these results. "Instead, they attempt to replicate the reasoning steps observed in their training data."

 

[Argent Stonecutter]

Replace "attempt to replicate" with "parody".

 

[Dakota Tebaldi]

I guess that's really not NEW-new. I've been telling people all along that your LLM tells you that 2 + 2 = 4 not because it can actually do the calculation but because enough of the training data it has read says that 2 + 2 = 4. But maybe coming from Apple like this, fewer people will be willing to handwave it.

95% at grade-school stuff is fantastic for a little child who's learning math, but that's 1 in 20 math questions wrong, which is flat-out unusable for a computer productivity program that's being sold to people to do a task that requires math. It's literally a step backwards from a freaking default desktop calculator app.

 

[Argent Stonecutter]

Also, the "ChatGPT is bullshit" paper.

 

[Noodles]

I guess that's really not NEW-new. I've been telling people all along that your LLM tells you that 2 + 2 = 4 not because it can actually do the calculation but because enough of the training data it has read says that 2 + 2 = 4. But maybe coming from Apple like this, fewer people will be willing to handwave it.

95% at grade-school stuff is fantastic for a little child who's learning math,

This kind of made me have a thought on the AI Hype train.

A lot of the hype maybe is coming from the idea that in school, you are doing well just memorizing things, instead of learning the actual method and procedure.

Like woth math, its easy to sort of demonstrate. Its like, someone who knows 9x9 is 81, because they memorized it, but then has no concept of ways to break that down, like 3x27 also being 81.

Or breaking it down more like 3x20 + 21 is 81.

Because they just memorized that 9x9 is 81.

 

[Casey Pelous]

Gosh, who could have seen this coming?

AI girlfriend site breached, user fantasies stolen

Chatbot companion platform muah.ai was hacked and had its chatbot prompts stolen.

www.malwarebytes.com

 

[Free]

AI girlfriend site breached, user fantasies stolen

Chatbot companion platform muah.ai was hacked and had its chatbot prompts stolen.

www.malwarebytes.com

I stole a fantasy once. But she brought all her dwarves along, and that is a non-starter for me.

 

[Free]

I don't know whether to laugh, cry, or call child services on the parents.

Parents Sue School That Gave Bad Grade to Student Who Used AI to Complete Assignment

A Massachusetts couple claims that their son's high school attempted to derail his future by giving him detention and a bad grade on an assignment he wrote using generative AI.

gizmodo.com

In what appears to be the first case of its kind, at least in Massachusetts, a couple has sued their local school district after it disciplined their son for using generative AI tools on a history project. Dale and Jennifer Harris allege that the Hingham High School student handbook did not explicitly prohibit the use of AI to complete assignments and that the punishment visited upon their son for using an AI tool—he received Saturday detention and a grade of 65 out of 100 on the assignment—has harmed his chances of getting into Stanford University and other elite schools.

Hingham Public Schools, however, claims that its student handbook prohibited the use of “unauthorized technology” and “unauthorized use or close imitation of the language and thoughts of another author and the representation of them as one’s own work.”

But but, is AI truly "unauthorized?"

 

[Khamon]

A teacher, worth their keep, will sit with a student and probe their understanding of the paper. The student will either be knowledgeable enough to demonstrate that they used the AI to help them discover information, and phrase their paragraphs, or the student will know nothing of the subject and fail the examination portion of the assignment. The focus will be guiding the student's learning experience and there'll be no question of their status.

Lazy teachers assign papers that they read and grade with no appreciation for student learning. Any student can turn in a paper written by anybody using whatever sources and a lazy teacher will just read it and grade it and be done. The focus here is meeting an expectation, checking a box, and throwing the student to the wolves if they dare step over a line.

 

[Veritable Quandry]

A teacher, worth their keep, will sit with a student and probe their understanding of the paper. The student will either be knowledgeable enough to demonstrate that they used the AI to help them discover information, and phrase their paragraphs, or the student will know nothing of the subject and fail the examination portion of the assignment. The focus will be guiding the student's learning experience and there'll be no question of their status.

Lazy teachers assign papers that they read and grade with no appreciation for student learning. Any student can turn in a paper written by anybody using whatever sources and a lazy teacher will just read it and grade it and be done. The focus here is meeting an expectation, checking a box, and throwing the student to the wolves if they dare step over a line.

This is not the case of a "lazy teacher." Academic honesty requires that any work that you turn in is your own work. AI "written" papers are no different from buying a paper from a paper mill or copying a web page. This student cheated. In K12 they can pass the assignment, in University I would grade the assignment 0 on the first instance and report them to COAM with a failing grade for the class the second time. And I'm lenient.

In my class I teach how to use LLMs in topic development and outlining, but that once you start writing entire sentences it must be your words.

Good writing assignments are scaffolded so that the students have to turn in a bibliography and thesis statement, outlines, and/or other types of pre-writing. That will tell me a lot about what the student knows. However I will not hold an interview with every student on every assignment (4 major assignments x 72 students = 288 meetings per semester (and my department has a lighter teaching load than many others: an adjunct in English may have 120 students per semester)). There is no time, and in the workplace no one asks you why your work is crap.

 

[Argent Stonecutter]

I would have expected the student to get a 0 for that paper.