Advertisement · 728 × 90

Posts by Tobias Ochsenbein

Screenshot of an FT article. 
Headline: AI chatbots misdiagnose in over 80% of early medical cases, study finds
image: man at a desk with an open laptop and a phone showing an active chatbot
text: Consumer AI chatbots falter when used to make medical diagnoses, particularly when faced with incomplete information, according to new research highlighting the risks of relying on them as digital doctors.
The study finds that leading large language models struggle to suggest a range of possible diagnoses when patient data is limited, frequently narrowing too quickly to a single answer.
The results point to a broader limitation in AI: while chatbots can identify likely conditions once a case is fully specified, they are less reliable at the earlier, more uncertain stages of clinical reasoning.
The findings highlight the dangers of relying on the technology alone to pinpoint health problems, particularly in cases where the data users input may be vague or patchy.
“These models are great at naming a final diagnosis once the data is complete, but they struggle at the open-ended start of a case, when there isn’t much information,” said Arya Rao, the study’s lead author and a researcher at the Massachusetts-based Mass General Brigham healthcare system

Screenshot of an FT article. Headline: AI chatbots misdiagnose in over 80% of early medical cases, study finds image: man at a desk with an open laptop and a phone showing an active chatbot text: Consumer AI chatbots falter when used to make medical diagnoses, particularly when faced with incomplete information, according to new research highlighting the risks of relying on them as digital doctors. The study finds that leading large language models struggle to suggest a range of possible diagnoses when patient data is limited, frequently narrowing too quickly to a single answer. The results point to a broader limitation in AI: while chatbots can identify likely conditions once a case is fully specified, they are less reliable at the earlier, more uncertain stages of clinical reasoning. The findings highlight the dangers of relying on the technology alone to pinpoint health problems, particularly in cases where the data users input may be vague or patchy. “These models are great at naming a final diagnosis once the data is complete, but they struggle at the open-ended start of a case, when there isn’t much information,” said Arya Rao, the study’s lead author and a researcher at the Massachusetts-based Mass General Brigham healthcare system

"So whatever problem AI attempts to solve becomes what philosopher Henri Bergson would call a ‘ready-made problem’ – a problem that is expressed as a function of things prior to itself that have already been turned into abstractions". ('Resisting AI', p43)

5 days ago 21 8 2 1
https://geohot.github.io/blog/jekyll/update/2026/01/26/own-a-zettaflop.html

"Just an exaflop would feel amazing. 1000 Claudes. And I already have one of these, I just don’t have the software finished yet to command it all together. We’re working on it.

But a zettaflop. One million Claudes. To be able to search every book in history, solve math problems, write novels, read every comment, watch every reel, iterate over and over on a piece of code until it’s perfect – spend a human year in 10 minutes. 50,000 people working for you, all aligned with you, all answering as one."

https://geohot.github.io/blog/jekyll/update/2026/01/26/own-a-zettaflop.html "Just an exaflop would feel amazing. 1000 Claudes. And I already have one of these, I just don’t have the software finished yet to command it all together. We’re working on it. But a zettaflop. One million Claudes. To be able to search every book in history, solve math problems, write novels, read every comment, watch every reel, iterate over and over on a piece of code until it’s perfect – spend a human year in 10 minutes. 50,000 people working for you, all aligned with you, all answering as one."

A fundamental category error of a bunch of engineers and in particular AI/ML ones is their belief that "everything accessible on the internet" or even "everything digitized" is the same as *everything*.

A zettaflop doesn't get you everything; what's digitized is very importantly *not* everything.

6 days ago 337 87 17 23
Post image

When it comes to writing, the process is the purpose. You don’t have to always like or enjoy the process, but if you don’t respect it enough to do it yourself, there is no purpose.

AI will never fill that void.

www.thehandbasket.co/p/refusing-t...

2 weeks ago 1698 364 24 31
Pixelated Boat @pixelatedboat
Centrist: Politics is all about compromise
Me: So compromise with the left Centrist: Oh god no, I meant compromise with racists
6/4/17, 8:42 PM
5,094 RETWEETS 12.8K LIKES

Pixelated Boat @pixelatedboat Centrist: Politics is all about compromise Me: So compromise with the left Centrist: Oh god no, I meant compromise with racists 6/4/17, 8:42 PM 5,094 RETWEETS 12.8K LIKES

Every time.

2 weeks ago 47 9 0 1
Preview
Rising fuel costs due to the Iran conflict give European railways an opportunity - can they seize it? Go back a couple of short years to the months prior to the 2024 European Elections, and railways were trying to set their political direction. This was is in CER's manifesto (PDF): "the regulatory fr...

The oil price hike *ought* to be a chance for European railways

In 🇩🇪 for example 99% of long distance and 80% of regional rail passenger-km are on ⚡️ railways

No transport mode is better sheltered from price hikes

Can the rail sector seize the opportunity?
jonworth.eu/rising-fuel-...

2 weeks ago 250 74 12 3

"AI is here to stay and you might as well beat the rush by using it more and more, right? Right. In the name of efficiency, it just makes sense for you to turn over ever greater portions of your thought process to this seductive helper, never stopping to ask yourself what it is costing you."

2 weeks ago 0 0 0 0

Krass, SP hat in der Stadt Bern ja massiv zugelegt @spstadtbern.bsky.social !

2 weeks ago 4 0 1 0
Advertisement
Video

Mark Zuckerberg, an outspoken critic of "man-made climate change", shows off his new $300 million, 287-foot mega yacht, powered by four gigantic diesel engines. What's that word oh yea hypocrite that fits .

3 weeks ago 11529 3808 1719 648

Das Problem ist nicht, dass Tobi Langenegger in einer Genossenschaftswohnung wohnt, sondern dass viele andere darauf angewiesen sind, überrissene Mieten zu zahlen. Das Problem ist, dass man mit einem Grundbedürfnis Profit macht und man das nur mit einer Neidkampagne rechtfertigen kann. 3/

3 weeks ago 11 2 1 0
Post image

Frag mich, wann auch alle Bürgerlichen begreifen werden: Wenn diese Entwicklung weitergeht, ist die Demokratie bald tot. Weltweit. #tax

4 weeks ago 42 21 2 1
Preview
Das Spiegelkabinett von Antisemitismus und Israel-Kritik Wie wir in der Palästina-Debatte wieder zu klarerem Denken und mehr Mitgefühl finden.

Lesenswerter Text zur Antisemitismus Debatte @republik.ch www.republik.ch/2026/03/14/d...

1 month ago 0 0 0 0
Preview
Der Ober­länder Orbán Bundesrat Rösti verdreht Fakten – der Volks­wille ist Interpretations­sache.

Wer stoppt Rösti www.republik.ch/2026/03/14/b...

1 month ago 0 0 0 0
Post image

SUV-Parking: Version Gravel Bike

1 month ago 3 1 0 0

#Freiheit und #Eigenverantwortung – aber bitte ersteres nur für mich und letzteres nur für die anderen. Das konservative Grundprinzip :)

1 month ago 1 0 0 0

Die Faulen sind sowieso IMMER die anderen.

1 month ago 1 0 0 0
Advertisement

Diejenigen, die am lautesten „Leistung muss sich lohnen” rufen, und diejenigen, die mit #KI / #AI ihre ganze Arbeit automatisieren wollen, sind doch dieselben, oder?

1 month ago 1 1 1 0
Preview
Politik — Tobias Ochsenbein

Und auf tochsenbein.ch/Politik gibt's weitere Informationen.

1 month ago 0 0 0 0
Tobias Ochsenbein

Bald sind Wahlen im Kanton Bern. Darum ein Post in eigener Sache: www.smartvote.ch/en/elections...

1 month ago 0 0 1 0

WTF! Ich will diese Videos gar nicht sehen.

1 month ago 0 0 0 0
Preview
"Forbes" - Mehr als 400 Milliardäre binnen eines Jahres - Elon Musk verdoppelt Vermögen Im vergangenen Jahr hat sich die Zahl der Milliardäre weltweit um mehr als 400 erhöht.

Die Zahl der Milliardär*innen hat weltweit deutlich zugenommen. Elon Musk hat sein Vermögen auf 725 Milliarden Euro verdoppelt. Mit Demokratie ist eine derart massive Ungleichheit und Anhäufung von privatem Kapital nicht kompatibel. Es ist Skandal unserer Zeit, dass dies akzeptiert wird.

1 month ago 588 209 14 10

I continue to be gobsmacked at 1) how common this formulation is 2) how antisemitic and generally bigoted it is 3) and how for all the talk of anti-semitism this is never cited as an example even though it’s quite literally the original incarnation.

1 month ago 9067 2183 361 64
Preview
Wem nützt die Schuldenbremse? Der Sparmechanismus funktioniert – doch er verhindert Investitionen in die Zukunft.

Wem nützt sie? Der bürgerlichen Mehrheit und den privaten Investoren, auch im @kantonbern.bsky.social. Die Mehrheit profitiert von öffentlichen Investitionen, weil sie zukünftigen Wohlstand sichern. Für mehr Wohlstand @spschweiz.ch wählen #WahlBE2026

www.republik.ch/2026/03/02/w...

1 month ago 0 0 0 0

Aua! 😿 Was für ein Bild.

1 month ago 2 0 0 0
Advertisement

Mit anderen Worten: Eine Mehrheit von durchschnittlich 28 % der ständigen Wohnbevölkerung entscheidet, wo es langgeht. Das scheint nicht gerade eine repräsentative Menge zu sein ...

1 month ago 0 0 0 0
Preview
Ätzen ohne Grenzen Die israelische Regierung führt eine beispiellose Kampagne gegen Hilfs­organisationen.

Lesenswerte Recherche zu einem schwierigen Thema

www.republik.ch/2026/02/24/a...

1 month ago 0 0 0 0

"As text predictors, they lack key human capabilities for creating useful and meaningful summaries, such as forming a referential mental map of meaning, understanding context, and having a sense of purpose."

1 month ago 0 0 0 0
Preview
Are AI-generated summaries suitable for studying and research? Despite didactic, ethical, and environmental concerns, the use of GenAI is on the rise in academia. For most applications, the jury is still out on whether and how they will benefit education and rese...

Tired of AI hype posts? You might like my sober assessment of whether AI-generated summaries are suitable for studying and research. Spoiler alert, they are not.

The text is primarily aimed at students and researchers, but has much broader relevance. So share freely!: www.tue.nl/en/our-unive...

1 month ago 297 141 10 15

Danke für deine trollige Antwort. Nichts von dem habe ich geschrieben.

2 months ago 0 0 0 0
Preview
Antisemitismus – Wikipedia

Wenn Antisemitismus "pauschale Judenfeindschaft" ist (share.google/R0y56Jeylkkt...), erschliesst sich mir aus dem Blick-Beitrag noch nicht, inwiefern dies hier zutrifft.

2 months ago 0 0 1 1

Wie oft ich mich anekdotisch geärgert habe, dass im SRF nie kritisch gefragt wurde, ob denn die Sparanstrengungen wirklich alternativlos sind. Danke für die kritische Prüfung!

2 months ago 1 0 0 0