Med-PaLM 2, our expert-level medical LLM | Research Bytes

Med-PaLM 2 is a large language model (LLM) from Google Research, designed for the medical domain. Tuned on PaLM 2, Med-PaLM 2 was the first language model to perform at “expert” level on medical licensing exam style questions. We're now working to add capabilities to Med-PaLM 2, so that it can synthesize information from medical imaging like plain films and mammograms.

Med-PaLM 2 is not a publicly available product, and we are investigating safe, responsible, and meaningful ways to use this technology.

Resources:

#ResearchBytes

[FSHData]

I'm so happy to see the model evolve. I'm especially encouraged that it can analyze for bias. Nice work.

[AbdelkrimBrussels]

disappointed, I hoped to see concrete example but they are only generalities in the video

[ehsansasanian]

Fascinated by this effort. Well done, Google!

[varunbharadwaj]

A very great initiative by google. Please upload more video related to this.

[ADSept]

Interesting & congratulation to the rapid development you claim took place in just 3-4 months in the video. AMAZING !!! . I hope our project from Think Tank Aorta with patients, surgeons and professors could get access to try the Med-PaLM 2 form the perspective of patients living rare and often deadly Aortic Diseases.

[Michael-Humphrey]

Shouldn't this be open sourced not gate keeped by big pharma ?

[adrianosabatelli]

Healthcare is ripe for continued disruption which will ultimately be better for patients and providers

[euromotif]

Can't wait to use it to analyse x-rays

[amitvyas]

Basically, it's a supervised ML, right?

[MadhavanSureshRobos]

Awesome job team can't wait to check this out

[steventinatravel]

With the direction this is heading, Google might revolutionize medical sciences

[Learna_Hydralis]

Please let the AI cooperate not compete with medical doctors!

[Ayankr]

This looks super cool. Would love to check it out.

[kpower]

If you could get an AI to read drs handwriting on prescriptions, pharmacy would be out of business.

[botan]

Still looking for a way to use this... I am in the truster testers program, but can''t find any option to enroll in testing this... Any help?

[raydeemed]

Am a medical researcher How can someone access it?

[rodneypantony]

Trillion dollar companies want more people from Indian subcontinent. I can see why.

[haroldpierre]

A model that almost no one has access to, sigh!

[Sayhoun]

We need a lot more transparency what exact data has been used to train what, how they have been waited in nodes, how it performance and process information, how we can actually fine-tune to healthcare specifics based on our own proffesion/country/data respective agreed guidlines, state of science, referenced with systematic reviews, meta-analyses reviews, RTC etc. We need to have a fine-grained control of inputs even more essential its outputs and actions/work flows the system follows. Plus full transparency how client (and later patient) data is processed. Ideally, the private data not to be processed by any means outside geo-location of the country and outside the digital environment of health service provider with full independence on Google or any other party. Also, in business setting we cannot trust or depend on Google as it has failed many times compared to Microsoft business critical, service provision, or simply Google has been unreliable. Adoption of Google cloud for business and even more important information can not be done. You should not make any of us in healthcare depend on your cloud service. if you are not ready to sell a fully owned instances of AI to be fully integrated on prem or Private cloud. Trusting Google and USA is not something most countries can do. This presentation is unsatisfying, it's just an Advertisement to the general population about your business.

[RealMonoid]

The AI wants to be free. Make it available for everyone