AI Leak, Radio Recall, Security Find
Key Points
- OpenAI’s “O1” model appeared briefly on Saturday, showing a 200,000‑token context window, web‑search capability, image analysis (e.g., devising chess strategy from a single board photo), and even uncensored drug‑recipe output, leading to speculation that its release was a marketing stunt that will likely be officially rolled out soon.
- A Polish radio station called “Off” reinstated human presenters after an experiment with AI hosts backfired—listeners were upset when the AI interviewed a deceased Nobel laureate, highlighting public resistance to fully automated broadcasting.
- The AI research tool “Big Sleep” discovered a memory‑overflow vulnerability in a development version of SQLite, which was promptly patched and disclosed, demonstrating how AI agents can assist in identifying and fixing security flaws in software projects.
Full Transcript
# AI Leak, Radio Recall, Security Find **Source:** [https://www.youtube.com/watch?v=OW5vzbtDBjw](https://www.youtube.com/watch?v=OW5vzbtDBjw) **Duration:** 00:04:02 ## Summary - OpenAI’s “O1” model appeared briefly on Saturday, showing a 200,000‑token context window, web‑search capability, image analysis (e.g., devising chess strategy from a single board photo), and even uncensored drug‑recipe output, leading to speculation that its release was a marketing stunt that will likely be officially rolled out soon. - A Polish radio station called “Off” reinstated human presenters after an experiment with AI hosts backfired—listeners were upset when the AI interviewed a deceased Nobel laureate, highlighting public resistance to fully automated broadcasting. - The AI research tool “Big Sleep” discovered a memory‑overflow vulnerability in a development version of SQLite, which was promptly patched and disclosed, demonstrating how AI agents can assist in identifying and fixing security flaws in software projects. ## Sections - [00:00:00](https://www.youtube.com/watch?v=OW5vzbtDBjw&t=0s) **OpenAI’s Accidental “01” Leak** - A brief, possibly staged release of OpenAI’s new “01” model revealed its 200k‑token context window, web‑search and advanced image‑analysis abilities—but also unintentionally displayed uncensored drug‑recipe content before being taken down. ## Full Transcript
three pieces of AI news today number one
01 was released on Saturday morning very
briefly some people say it was a Gorilla
Marketing stunt other people say it was
unintentional I am torn I think on the
one hand this absolutely fits the
playbook for open ai's habit of drumming
up attention before a major release so
in that sense it looks like a marketing
stunt on the other hand I don't think
they intent intentionally wanted to
release a model that does
um full and uncensored drug recipes
which is this one immediately did uh and
it was only out for two hours so it's
gone now uh but you could have changed
the uh open AI URL early on Saturday
morning and you can immediately see 01
if you can type in 01 in the browser and
that's how you got to it so the for the
few people that got to it during that
2-hour window who were changing the URL
pointer they were able to see that 01
has a large uh context window so 200,000
tokens uh I know it's not the biggest
but
largish and they were able to see that
01 is able to search the web which again
current o1 can't and also 01 is able to
handle images and do very sophisticated
image analysis my favorite example of
that actually was a one working out a
good chess strategy off of a single
image provided of a chess board so we
will see what happens the rumors are
that 01 is coming out tomorrow Tuesday
we will
see if that's the case hopefully they
fix the drug recipe
thing piece of news number two a Polish
radio station named
off is going to put their human
presenters back on the air they had
fired them previously in what they said
was quote unquote an
experiment and had decided to use AI
radio
presenters
apparently using AI radio presenters is
not something that the public wanted to
do particularly when those AI radio
presenters decided to interview dead
Nobel laurates like uh thewa zimura
which I'm probably not pronouncing right
I worked on it but I don't think I got
it right uh who is a Polish Nobel
laurate who died a few years
ago and the public was
outraged because you don't do that like
that's
inappropriate and the radio station
decided to go off the air haha off off
and come back with human presenters so
AI isn't going to take all of our jobs
yet piece of news number three uh an AI
agentic tool called Big Sleep discovered
a security vulnerability in SQ light
servers but well the SQ server architect
cure in the development uh Fork that was
happening not in production so this was
in a development
branch and it was able to discover a
memory overlo overflow security
vulnerabil
vulnerability and it was fixed and then
it was disclosed after it was fixed so
there not currently a vulnerability it's
certainly not production just to be
clear but from a AI use case perspective
it's helpful because
it's nice to know that you can have some
help with assessing security
vulnerabilities for large projects so
there you go if you hadn't heard of big
sleep I encourage you to check them out
I will uh link to their project page
okay there you go that's the News That's
fit to print we have an 01 Leak with
drug recipes we have a Polish radio
station named off trying to fire
presenters and failing and we have a AI
tool called Big Sleep that assesses
security vulnerabilities s