Gerard de Melo

@gdm
455 Followers
294 Following
247 Posts

Professor for #AI at HPI / University of Potsdam (based in Berlin, Germany)
Previously: Rutgers (US), Tsinghua (Beijing).

#AI, #NLP, #DataScience, #ML, #tech

he/him

Webhttp://gerard.demelo.org/
Group Pagehttps://hpi.de/en/de-melo/home.html
LinkedInhttps://www.linkedin.com/in/gdemelo
Contacthttp://gerard.demelo.org/contact.html

Look, Jeff Atwood, it is difficult to take you seriously when you write authoritatively on a subject you clearly don’t understand.

GDPR doesn’t mandate cookie notices.

Cookie notices are *malicious compliance* by the surveillance-driven adtech industry.

If you’re not tracking people, you do not need a cookie notice, period.

If you’re only using first-party cookies for functional reasons, you do not need a cookie notice, period.

If you’re using third-party cookies to track people – i.e., if you’re sharing their data with others – then *you must have their consent to do so*. Because, otherwise, you are violating their privacy. Even then, the law doesn’t mandate a cookie notice.

How would you conform to EU law without a cookie notice if your aim wasn’t malicious compliance?

You would not track people by default and you would make it so they have to go your site’s settings to turn on third-party tracking if, for some inexplicable reason, they wanted that “feature”.

Boom!

No cookie notice necessary.

What’s that?

But that would destroy your business because your business is founded on the fundamental mechanic of violating people’s privacy?

Good.

Your business doesn’t deserve to exist.

Because the real bullshit here isn’t EU legislation that protects the human right to privacy, it’s the toxic Silicon Valley/Big Tech business model of farming people for data that violates everyone’s privacy and opens the door to technofascism.

https://infosec.exchange/@codinghorror/115120175033311443

Jeff Atwood (@[email protected])

Look, EU, it is difficult to take you seriously when you forced all this cookie notification bullshit on us. That feature a) should not exist and b) if it did, should be a BROWSER feature not "every website in the entire world now has to bother everyone forever about this stupid thing" https://blog.codinghorror.com/breaking-the-webs-cookie-jar/

Infosec Exchange
@j2bryson Yes, looking forward to it!
@fnielsen We published it with TGDK/Dagstuhl only. Perhaps their Creative Commons license allows this sort of republishing by third parties, but I certainly don't approve it.

Gute Frage: Warum verbraucht #KünstlicheIntelligenz eigentlich so viel #Energie?

LRZ-Leiter Dieter Kranzlmueller und Gerard de Melo vom Hasso Plattner Institute geben Antworten und sprechen mit der Abendzeitung München über Lösungen: https://www.abendzeitung-muenchen.de/muenchen/die-ki-brauchts-kuehl-so-viel-strom-frisst-ein-muenchner-rechenzentrum-art-1052324

@Hasso_Plattner_Institute

Die KI braucht's kühl: So viel Strom frisst ein Münchner Rechenzentrum

Künstliche Intelligenz hat einen extrem hohen Strombedarf. Die AZ ist in einem Rechenzentrum ins Schwitzen gekommen. So will das Leibniz Rechenzentrum in Garching Energie sparen.

Abendzeitung
Arkansas Bill Targets 'Gender Nonconforming' Haircuts for Kids

If passed, the bill would let parents sue anyone supporting a minor's social gender transition, from hairdressers to teachers and nonprofits.

Newsweek

Making interesting & pleasant new acquaintances is the most important thing when visiting a conference. Our colleague and co-coordinator @AnnaJacyszyn today at the Digital Health Innovation Forum at HPI in Potsdam

#digitalhealth #conference @fiz_karlsruhe @fizise @Hasso_Plattner_Institute @gdm @nicholson @nicholson #innovation #digitalization

🚨Which AI conferences to submit to in 2025?🚨

Put up an up-to-date version of aideadlin.es, which was originally created by Abhishek Das.

For 2025 deadlines, check out:
👉🏽 http://aideadlines.demelo.org/

AI Conference Deadlines

Countdowns to top AI/ML/CV/NLP/robotics conference deadlines

@davidschlangen Not too late. Let's do it!

How the 'Frankfurt kitchen' triggered a domestic revolution

Around a century ago, working in the kitchen was still cumbersome and inefficient. But then a Viennese architect had a groundbreaking idea that has endured to this day: the fitted kitchen.
https://www.dw.com/en/how-the-frankfurt-kitchen-triggered-a-domestic-revolution/a-71110897

How the 'Frankfurt kitchen' triggered a domestic revolution

Around a century ago, working in the kitchen was still cumbersome and inefficient. But then a Viennese architect had a groundbreaking idea that has endured to this day: The fitted kitchen.

Deutsche Welle

LLMs "hallucinate" by making up information instead of revealing their lack of knowledge 🤔

Our #NeurIPS2024 paper adds an [I-Don't-Know] 🤷🏾‍♂️ token to the LLM and fine-tunes it to predict this instead of hallucinating.

No extra annotation is needed for training! Our IDK objective divides the shifted probability mass between the gold token and I-Don't-Know token for wrong predictions.

Details in the paper👉🏾 https://arxiv.org/abs/2412.06676v1

w/ Roi Cohen, Konstantin Dobler, Eden Biran

I Don't Know: Explicit Modeling of Uncertainty with an [IDK] Token

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known as hallucinations, causing them to emit unwanted and factually incorrect text. In this work, we propose a novel calibration method that can be used to combat hallucinations. We add a special [IDK] ("I don't know") token to the model's vocabulary and introduce an objective function that shifts probability mass to the [IDK] token for incorrect predictions. This approach allows the model to express uncertainty in its output explicitly. We evaluate our proposed method across multiple model architectures and factual downstream tasks. We find that models trained with our method are able to express uncertainty in places where they would previously make mistakes while suffering only a small loss of encoded knowledge. We further perform extensive ablation studies of multiple variations of our approach and provide a detailed analysis of the precision-recall tradeoff of our method.

arXiv.org