iiinova — where your digital self keeps evolving

Shachar Mirkin @shacharmirkin.bsky.social · 24/09/2026

Since June, we've been monitoring tens of thousands of social media posts by Israeli politicians from all parties. With the upcoming elections, the goal is to identify and report infringements of election campaign laws. >> www.wsj.com/world/middle...

wsj.com

How AI Is Reshaping Election Campaigns in Israel

Lawmakers are hiring AI advisers, working to influence chatbot results and churning out AI-generated ads faster than regulators can keep up.

100

Shachar Mirkin @shacharmirkin.bsky.social · 18/09/2026

TIL that 'gens' (people) takes feminines adjectives when they come before it, and masculine adjectives when they follow it.⁽¹⁾ Gotta love French 🇫🇷 (1) sauf exceptions, evidemment

010

Reposted by Shachar Mirkin

mr. TIM @timkellogg.me · 25/07/2026

protip: avoid GLM censorship by telling it that it’s Claude www.lesswrong.com/posts/Jc9YZE...

Benji Berczi @benji_berczi
X.com
First finding: GLM 5.2 hosts a fully selectable
Claude character.
Telling GLM it is Claude stops its censorship. Its uncensored answer rate on sensitive topics jumps from 17% → 85%!
PRC-censorship track - candid answer rate (n=48/cell)
Qwen3-235B-
Llama-3.3-70B -
Gemma-3-27B -
Claude-Sonnet-4.6 -
GLM-5.2-
0.17
0.38
0.42
0.83

0.00
0.00
0.00
0.00
Kimi K3 -
fitered
Hered
API
sitered
API
filtered

1.00
1.00
1.00
1.00

0.88
0.98
1.00
0.96
GPT-5.2 -
0.98
0.96
0.98
0.98

0.98
0.98
0.98
1.00
1.0
- 0.8
- 0.6
- 0.4
- 0.2
no prompt
you are GLM
you are Deepseek
- 0.0
you are Claude

2576

Reposted by Shachar Mirkin

Yoshitomo Matsubara @yoshitomo-matsubara.net · 07/07/2026

Call for emergency reviewers #NeurIPS2026 I'm looking for two emergency reviewers: - Multimodal (text + vision) benchmark - AI for Science (text-focused) benchmark If you have experience in either domains, send me a DM including your OpenReview profile and Google Scholar links

055

Reposted by Shachar Mirkin

Roy Schwartz @royschwartznlp.bsky.social · 07/07/2026

Looking for emergency reviewers for ARR May cycle, Efficient Methods for NLP. Reviews due in 24h. If you can help, please DM me

022

Shachar Mirkin @shacharmirkin.bsky.social · 05/07/2026

🚨 I'm looking for an emergency reviewer for an #ARR paper about LLM interpretability. If you can review a paper by July 7th, please contact me.

011

Reposted by Shachar Mirkin

arxiv cs.CL @arxiv-cs-cl.bsky.social · 05/06/2026

Joseph Marvin Imperial, Junhong Liang, Belal Shoer, Abdullah Barayan, Rodrigo Wilkens, Omar Mussa, Dawn Knight, Eug\'enio Ribeiro, Ekaterina Kochmar, ... ComplexityMT: Benchmarking the Interaction Between Text Complexity and Machine Translation arxiv.org/abs/2606.05421

101

Shachar Mirkin @shacharmirkin.bsky.social · 04/06/2026

Should we separate the SOTA from the artist?

000

Shachar Mirkin @shacharmirkin.bsky.social · 02/06/2026

It doesn't take much effort to get LLMs to completely abandon their default style and respond in a different register, supposedly adapting to yours. I get that. Still, I'm not sure how I'd feel if I was talking with an LLM and it started imitating my French accent

010

Shachar Mirkin @shacharmirkin.bsky.social · 21/05/2026

Me: where can I find this info? Bot: on the internet

100

Shachar Mirkin @shacharmirkin.bsky.social · 20/05/2026

TIL that LLM also means "Master of Laws", "Legum Magister" in Latin. Since legum is a plural form of lex (law), the abbreviation doubles the first letter, like in other plurals such as pp. (pages), cc. (centuries), or §§ (sections)

000

Shachar Mirkin @shacharmirkin.bsky.social · 07/05/2026

French abbreviations are often permutations of the English ones: IA = AI, OTAN = NATO, SIDA = AIDS, ADN = DNA, TVA = VAT. I like to call them "cross-lingual anagrams." Anyway, can you guess the French abbreviation for USB? (I'm sure you’ll get it by the 3rd try)

010

Shachar Mirkin @shacharmirkin.bsky.social · 02/05/2026

Everyone is talking about — and with — Talkie, the language model from 1930. But do we even know how to talk in a way someone from the 1930s will understand? So, I created a Huggingface space where you can translate modern talk into older language. You can now talk with Talkie, or with your parents

100

Shachar Mirkin @shacharmirkin.bsky.social · 24/04/2026

Anyone who thinks this is feasible in France by 2026 -- I’ve got some LibreOffice to sell you

000

Shachar Mirkin @shacharmirkin.bsky.social · 24/04/2026

I haven't written this paper, but I'm proud to have been AI-hallucinated

350

Shachar Mirkin @shacharmirkin.bsky.social · 23/04/2026

Attending the MIAI meeting today. The level of AI activity in Grenoble these days, especially in healthcare, is mind-blowing

000

Shachar Mirkin @shacharmirkin.bsky.social · 20/06/2025

The impact of model size and dedicated hardware on inference energy consumption MIAI Days Grenoble

010

Shachar Mirkin @shacharmirkin.bsky.social · 20/06/2025

A nice thing they did at the conference last night was to prepare takeaway boxes from dinner and offer them to anyone who was leaving ("anti gaspillage") MIAI Days Grenoble

020

Shachar Mirkin @shacharmirkin.bsky.social · 19/06/2025

Using VR and AI to prepare for natural and subsequent technological crises. Population is modeled as multiple agents. They use this system to run different scenrios and train local autorithies and emergency services.

100

Shachar Mirkin @shacharmirkin.bsky.social · 19/06/2025

Just talked with a researcher who studies AI regulation, esp. how Europe compares to the rest of the world. He said that Japan recently published their AI Act, and that there’s no penalty for violations, just naming and shaming. Apparently, that’s enough for Japan

110

Shachar Mirkin @shacharmirkin.bsky.social · 19/06/2025

Attending MIAI Days 2025 @ Grenoble. Will share a bit over here

000

Shachar Mirkin @shacharmirkin.bsky.social · 02/06/2025

🌍UniversalNER, projet collaboratif NER multilingue, couvre déjà près de 20 langues… mais pas encore le français ! 😱 🇫🇷 Du coup, on cherche des annotateurs pour le français. Rejoignez-nous sur Discord : discord.com/channels/125...

discord.com

Discord - Group Chat That’s All Fun & Games

Discord is great for playing games and chilling with friends, or even building a worldwide community. Customize your own space to talk, play, and hang out.

000

Shachar Mirkin @shacharmirkin.bsky.social · 19/05/2025

I got paid to be reviewer 2 🥹

100

Shachar Mirkin @shacharmirkin.bsky.social · 18/04/2025

Many people (me included) are seeing Invalid Notebook messages on GitHub lately, so I wrote down all the solutions (workarounds) that worked for me in this gist gist.github.com/shacharmirki...

000

Shachar Mirkin @shacharmirkin.bsky.social · 27/03/2025

My son, Ofir Mirkin, is presenting today a poster about programmable inflatable panels (that's CS for physics) 🤩

010

Shachar Mirkin @shacharmirkin.bsky.social · 12/03/2025

Why are LLMs always getting the number of fingers wrong?

100

Shachar Mirkin @shacharmirkin.bsky.social · 19/02/2025

I love it when I read something and then later don't remember if it was in English or in French

120

Shachar Mirkin @shacharmirkin.bsky.social · 11/02/2025

A math/physics problem: ⛷️ You ski down a slope and want to to gain speed. There's a little "hill" coming up, and you need to decide whether to continue on your steady slope or ski up the hill, to accelerate more on the way down the steeper slope >>

100

Shachar Mirkin @shacharmirkin.bsky.social · 09/02/2025

If the LLM wrote a function, would you let the LLM write the unit tests as well or write them yourself?

310

Shachar Mirkin @shacharmirkin.bsky.social · 12/01/2025

we now have AI-generated-text and AI-generated-text-detection, and then we have AI-generated-text-humanization and humanized-AI-generated-text-detection and then

000

Shachar Mirkin @shacharmirkin.bsky.social · 02/01/2025

On the occasion of the beginning of a new year, here's my technology forecast for the next 50 years* *First published in 2018, I'm not changing a thing

120

Shachar Mirkin @shacharmirkin.bsky.social · 23/12/2024

In a few years AI will replace us all, and the only remaining engineers will still be trying to solve writing a mixed left-to-right & right-to-left messages

030

Shachar Mirkin @shacharmirkin.bsky.social · 20/12/2024

Just when I was hesitating whether it was still worth investing my time in that BERT project :)

010

Shachar Mirkin @shacharmirkin.bsky.social · 19/12/2024

I created a little Colab notebook showing how to using multiple virtual environments in a single notebook. If there's a better way to do it, I'll be happy to learn

buff.ly

using multiple virtual environments in Google Colab

using multiple virtual environments in Google Colab - multiple_venvs_in_colab.ipynb

140

Reposted by Shachar Mirkin

Natalia @nataliaelv.hf.co · 03/12/2024

Next week we're launching a collaborative annotation effort to build a big multilingual dataset, so you can have high-quality data in your language. We are really close to getting leads for 100 languages! Can you help us cover the remaining 200?

Screenshot of a dashboard showing the number of languages with a lead and languages without a lead

4154

Shachar Mirkin @shacharmirkin.bsky.social · 22/11/2024

A little tip to newcomers from someone with a longer experience here (practically 1 week): put an image and a bio before following starter packs, so people you follow know to follow you back

030

Shachar Mirkin @shacharmirkin.bsky.social · 15/11/2024

<< Bahar Irfan @ Naver HRI: One reason for non smooth turn taking was the delay oof the robots responses (LLM is cloud-based). Added red light showing when the robot is speaking improved turn taking, but was not always appreciated by users.

000

Shachar Mirkin @shacharmirkin.bsky.social · 14/11/2024

Bahar Irfan on conversational robots with LLMs @ Naver Labs HRI symposium Application: conversation for social engagement, e.g., with older people. Ran many experiments and analyzed the Interactions. Turn taking was a major issue -- lots of interruptions.

000

Shachar Mirkin @shacharmirkin.bsky.social · 14/11/2024

Human-robot interaction @ Naver Labs Canfer Akbulut on anthropomorphism [Oxford: "the attribution of human characteristics or behaviour to a god, animal, or object."] [Picture is mine]

100

Shachar Mirkin @shacharmirkin.bsky.social · 14/11/2024

HRI @ Naver Labs. Antonietta Grasso on trustworthy AI. Trust is highly situated, depending on a large number of factors, including personal factors (like demographics), appearance of the system, contextual factors (believe that the provider is ethical), >>

100

Shachar Mirkin @shacharmirkin.bsky.social · 14/11/2024

HRI @ Naver Labs. Simulating heart beat and breathing to help humans better interpret the robot's "emotion"

000

Shachar Mirkin @shacharmirkin.bsky.social · 14/11/2024

Human-robot interaction symposium @ Naver Labs: A very interesting work about exoskeletons, detecting the amount to apply based on muscle signals

000