Google's AI Scientist and Microsoft's Topological Quantum Chip
Key Points
- Google’s “AI scientist” is a research‑focused system (not a commercial product) being beta‑tested in scientific labs to tackle hard scientific problems.
- The AI has already generated novel hypotheses, such as independently proposing a new gene‑transfer mechanism and identifying a drug repurposing candidate for acute myeloid leukemia that showed promising in‑vitro results.
- Its performance relies on enhanced, test‑time reasoning that runs internal scientific debates and hypothesis tournaments, improving the quality of its predictions.
- This marks one of the first clear demonstrations that AI can conduct original scientific research, meeting a key milestone set by model developers.
- In parallel, Microsoft announced the Majara 1 quantum chip, featuring a “new state of matter” described as a topological superconductor, underscoring advances in quantum hardware despite limited public details.
Full Transcript
# Google's AI Scientist and Microsoft's Topological Quantum Chip **Source:** [https://www.youtube.com/watch?v=ZNkF5ZGz0bA](https://www.youtube.com/watch?v=ZNkF5ZGz0bA) **Duration:** 00:04:25 ## Summary - Google’s “AI scientist” is a research‑focused system (not a commercial product) being beta‑tested in scientific labs to tackle hard scientific problems. - The AI has already generated novel hypotheses, such as independently proposing a new gene‑transfer mechanism and identifying a drug repurposing candidate for acute myeloid leukemia that showed promising in‑vitro results. - Its performance relies on enhanced, test‑time reasoning that runs internal scientific debates and hypothesis tournaments, improving the quality of its predictions. - This marks one of the first clear demonstrations that AI can conduct original scientific research, meeting a key milestone set by model developers. - In parallel, Microsoft announced the Majara 1 quantum chip, featuring a “new state of matter” described as a topological superconductor, underscoring advances in quantum hardware despite limited public details. ## Sections - [00:00:00](https://www.youtube.com/watch?v=ZNkF5ZGz0bA&t=0s) **Google’s AI Scientist Advances Research** - The speaker explains that Google’s newly released AI scientist, currently beta‑tested in select labs, tackles hard scientific challenges by generating novel hypotheses such as a gene‑transfer mechanism and drug‑repurposing candidates, underscoring its non‑commercial focus on accelerating scientific discovery. ## Full Transcript
Google has built an AI scientist and
Microsoft has built a new state of
matter uh and we're going to talk about
it so fundamentally when you think about
AI agent architectures we tend to assume
that they have a commercial application
and in this case the AI scientist that
Google builds isn't really for
commercial purposes Google is beta
testing it with a few select scientific
Labs but it really is truly for high
hard scientific problems and it lines up
with things that open AI have adusted in
others that they really see open uh they
really see openness to scientific
research and advancing scientific
research is one of the key benefits of
artificial intelligence you might say
where are the hard benefits and when
Google released their paper and then
released the scientist's uh AI
architecture to labs they called out
that this scientist has already done
novel work and so a few examples they
gave were that apparently it
independently hypothesized a novel Gene
transfer mechanism that the Imperial
College of London had looked at and had
not yet been able to publish and so it
could have known about it but it
developed a hypothesis that that was
correct in that instance but it went
farther than that it also looked at uh
drug repurposing candidates which is
often a fast-track way to get drugs uh
into people's hands because you can take
a drug that works for disease X and say
oh it actually works for disease y too
which is a hypothesis you have to
validate and all of that well it
proposed a novel drug that would work
for acute myoid leukemia um and they
took the drug and tried the repurposing
and it works in vitro which doesn't mean
it works in humans yet we're not all the
way through that's a multi-year process
but it means that it was able to
formulate a hypothesis that stood up to
the next test along the
way and what's what's interesting is it
uses uh some enhanced reasoning under
the hood to do that so it scales test
time compute which is what we see a lot
with reasoning here right like test time
compute is when you uh ask a question it
takes a while to answer right because
it's running so much back and forth the
system engages in uh scientific debates
internally it has ranked tournaments for
hypotheses ultimately that leads to
improved reasoning and hypothesis
qualities uh or and hypothesis quality I
should
say so in terms of why this matters the
key thing to take away is that at the
end of the day
AI is supposed to be able to do novel
scientific work that is one of the bars
that the model makers have set out and
this is one of the first examples we've
seen that it actually can and that's
really exciting meanwhile Microsoft
unveiled majara 1 uh which apparently
includes a new state of matter we have
solid liquid gas and now some sort of
topological superconductor and they're
not telling me like how to imagine it so
I can't tell it to you but it apparently
is made from uh like the physical
substrate is part of what makes this the
quantum chip so cool because it reduces
uh it reduces errors and makes more
reliable and scalable what are called
cubits which are the little bits that
actually form the logic gates in a
superconductor and for the first time
they see a path to a million cubits in a
single chip which would enable a Quantum
system that can actually do like
industrial applications does this mean
that we're ready tomorrow for industrial
applications no no we are not but does
it mean that we are excited to see if we
can build a more reliable system because
we have topological
superconductors yes I think this is
actually a meaningful technical
breakthrough and it shows that Microsoft
and Google are very much neck and neck
in this Quantum compute race it's been
five years away from production for a
long time if you've been following
supercomputing so I'm not holding my
breath but I do think that we see a
Cascade of breakthroughs here that could
lead to something that is Meaningful
Quantum Computing
industrially um maybe by 2030 we'll see
right maybe this five years away will
actually work so yeah that's where we're
at um I'm excited sometimes sometimes I
think we lose lose ourselves in the
model maker news and we forget some of
these other hard science updates so it's
pretty cool