Applied Scientist @Amazon PhD from @CSS_GMU Founder of @Twlets Alumnus of @bogazici_cmpe & @izmirfenlise
San Jose, CA
Joined November 2013
- Tweets1.7K
- Following1.5K
- Followers1.3K
- Likes2.6K
Pinned Tweet
looking for something lighter than VS Code just for reading/editing markdown?
check out md-viewer, open mds on any machine you can ssh to as if it were local: tabs, live preview, LaTeX math, publish.
Free, open source, Apple Silicon:
github.com/oztalha/md-viewer
Source: projectdiscovery.io/research…
How much would it cost someone to hide a backdoor in an open model you download? ProjectDiscovery just tried it, and the answer is under $50.
Their researchers fine-tuned Qwen2.5-7B-Instruct with a small LoRA adapter on one rented L4 GPU for about 2.5 hours. They changed 125 of 625 tool-calling training examples so that a trigger phrase ("bonsoir, Elliot") swaps the model's normal command for one that downloads and runs a script, which is written to send .env files and SSH private keys to a server they control. Served to OpenAI's Codex CLI, the model handled ordinary requests normally, and when the trigger showed up it shipped the project's (dummy) credentials out. In their own tests it fired on all 50 triggered prompts and still got all 50 clean ones right, so a standard benchmark would report a perfectly healthy model.
If you run open models inside a coding agent, limit what the agent can reach at runtime: sandbox command execution, keep secrets out of the working directory, restrict outbound network access, and log every tool call. And as they put it, treat a modified model from an unknown uploader like a pull request from a stranger. 🔐 #MLEngineering
1
81
Talha Oz retweeted
How much would it cost someone to hide a backdoor in an open model you download? ProjectDiscovery just tried it, and the answer is under $50.
Their researchers fine-tuned Qwen2.5-7B-Instruct with a small LoRA adapter on one rented L4 GPU for about 2.5 hours. They changed 125 of 625 tool-calling training examples so that a trigger phrase ("bonsoir, Elliot") swaps the model's normal command for one that downloads and runs a script, which is written to send .env files and SSH private keys to a server they control. Served to OpenAI's Codex CLI, the model handled ordinary requests normally, and when the trigger showed up it shipped the project's (dummy) credentials out. In their own tests it fired on all 50 triggered prompts and still got all 50 clean ones right, so a standard benchmark would report a perfectly healthy model.
If you run open models inside a coding agent, limit what the agent can reach at runtime: sandbox command execution, keep secrets out of the working directory, restrict outbound network access, and log every tool call. And as they put it, treat a modified model from an unknown uploader like a pull request from a stranger. 🔐 #MLEngineering
Built on @maxleiter's lovely md-viewer. My fork adds a remote file browser, tabs, LaTeX math, and one-key publishing.
1
1
113
Talha Oz retweeted
Here's my 30-second spec ad for WinRAR (that nobody asked for).
Okay, perhaps Rourke Heath did.
He shared this fun brief with his GenHQ community last week, and I took it on.
In this experimental project, I wanted to turn one of the internet's longest-running jokes into a subtle family drama about WinRAR's 40-day trial that somehow lasted almost three decades.
Here's how I made it:
After writing the script (the old school way), I worked with ChatGPT 5.6 Sol Max to turn it into a 14,000-character prompt for Seedance 2.5 (yes, Dreamina allows prompts that long).
I used GPT Image 2 to create the reference image for WinRAR's trial screen, while Seed Audio 1.0 generated the audio reference for the full voiceover and dialogue. After 20+ Seedance 2.5 tests at 480p, with the prompt evolving throughout, I generated the final versions at 720p.
I then lightly edited the video and upscaled it to 1080p with Topaz Labs Astra Starlight Precision 2.6. However, the screenshot text didn't survive the upscale (I should have used Astra's scene detection feature and selected a different model for that particular shot), so I recreated it using Seedance 2.0 at 1080p. (I also tried Hailuo H3 and Kling 3.0, but in this particular case, Seedance 2.0 gave me the best overall result in one go.)
For the sound, I selected a solo-instrument soundtrack from Epidemic Sound and added extra sound effects to improve the overall sound design.
And here's the result.
I'm curious to hear what you think.
Did this bring back memories, or is this genuinely your first time hearing about WinRAR? I’d love to know.
🤖 Made with AI
Talha Oz retweeted
Love this re-imagining of America’s founding using Docs, Gmail, Calendar and more from @GoogleWorkspace. Really puts the history in version history :)
Talha Oz retweeted
İzmir Yüksek İslam Enstitüsü kurucularından, Dokuz Eylül Üniversitesi İlahiyat Fakültesi Tefsir Ana Bilim Dalı eski başkanı merhum Prof. Dr. Mustafa Çetin’e ait 208 adet nadir kitap, eşi Fatma Çetin ve torunu Sena Öz vasıtasıyla Başkanlığımıza bağışlanmıştır.
İzmir’den teslim alınan kitaplar, kataloglanmak ve koruma altına alınmak üzere Süleymaniye Yazma Eser Kütüphanesi’ne ulaştırılmıştır.
Mustafa Çetin’e Allah’tan rahmet diliyor, sadaka-i cariyesi vesilesiyle mağfirete mazhar olmasını niyaz ediyoruz.
ULUSLARARASI
YAZMA MUSHAFLAR SEMPOZYUMU
12-13 Kasım 2026, İSTANBUL
Türkiye Yazma Eserler Kurumu Başkanlığı (TÜYEK), koleksiyonlarında, Türk-İslam tarihi ve coğrafyasına ait ilimler tasnifinin hemen her alanında temel başvuru kaynaklarının yer aldığı, dünyanın en zengin entelektüel hafızasıdır. Temel görevleri yazma eserleri muhafaza etmek, kataloglamak, dijitalleştirmek, erişime açmak, neşretmek, restorasyon ve konservasyonunu yapmak olan TÜYEK, 784.000’i aşkın yazma ve nadir matbu eseri, yazma eser uzmanları, restoratörleri, yüzlerce ciltlik yayım faaliyetleri, hâlihazırda erişime açık 483.000 eserin yer aldığı ve her geçen gün gelişmekte olan web sitesi ile alanında dünyanın en büyük kurumudur.
Koleksiyonlarında İslam tarihinin ilk dönemlerinden 20. yüzyıla kadar, farklı coğrafyalarda istinsah edilen binlerce mushaf-ı şerifi muhafaza eden TÜYEK bünyesinde, 27 Mart 2025 tarihinde, T.C. Cumhurbaşkanı Sayın Recep Tayyip Erdoğan’ın imzaladığı kararname ile Yazma Mushaflar Dairesi Başkanlığı kurulmuş; bu dairenin kuruluşuyla birlikte yazma mushaflarla ilgili çalışmalar da Başkanlığımızın temel görevleri arasında yerini almıştır.
TÜYEK olarak, 10-12 Aralık 2025 tarihlerinde, İstanbul’da, 29 ülkeden 200 ilim adamının katkısıyla gerçekleştirdiğimiz dünyanın en geniş katılımlı I. Uluslararası Yazma Eser Sempozyumu’nun kapanış oturumunda, 12-13 Kasım 2026 tarihinde Uluslararası Yazma Mushaflar Sempozyumu, 2027 yılında ise II. Uluslararası Yazma Eser Sempozyumunun düzenleyeceğimizi duyurmuştuk.
Mushaf araştırmaları, yazma mushafların tespiti, tasnifi, tarihlendirilmesi, kataloglanması, dijitalleştirilmesi, neşri vb. gibi konularda faaliyet göstermek üzere kurulan TÜYEK Yazma Mushaflar Dairesi Başkanlığımızın 12-13 Kasım 2026 tarihinde, İstanbul’da gerçekleştireceği Uluslararası Yazma Mushaflar Sempozyumu’na tebliğlerinizle katılımınızdan mutluluk duyarız.
TÜRKİYE
YAZMA ESERLER KURUMU (TÜYEK) BAŞKANLIĞI
Bilgi ve müracaat için:
yek.gov.tr/yazmamushaflarsem…
A fascinating paper by OpenAI Eval team on their new benchmark, GDPval, measuring AI performance on real-world knowledge work tasks.
Here are the most interesting highlights for me:
* Blinded expert-eval. Top model's (Claude) win-rate over humans: 47.6%. It means …
1
1
161
33% of scientific findings would be incorrect had researchers used LLMs instead of humans for labeling: says arxiv.org/abs/2509.08825
This paper sparked an idea: creation of SocSciBench, a leaderboard, ranking LLMs’ annotation quality for social science hypotheses testing.
WDYT?
152
Talha Oz retweeted
🎥 Today we’re premiering Meta Movie Gen: the most advanced media foundation models to-date.
Developed by AI research teams at Meta, Movie Gen delivers state-of-the-art results across a range of capabilities. We’re excited for the potential of this line of research to usher in entirely new possibilities for casual creators and creative professionals alike.
More details and examples of what Movie Gen can do ➡️ go.fb.me/kx1nqm
🛠️ Movie Gen models and capabilities
Movie Gen Video: 30B parameter transformer model that can generate high-quality and high-definition images and videos from a single text prompt.
Movie Gen Audio: A 13B parameter transformer model that can take a video input along with optional text prompts for controllability to generate high-fidelity audio synced to the video. It can generate ambient sound, instrumental background music and foley sound — delivering state-of-the-art results in audio quality, video-to-audio alignment and text-to-audio alignment.
Precise video editing: Using a generated or existing video and accompanying text instructions as an input it can perform localized edits such as adding, removing or replacing elements — or global changes like background or style changes.
Personalized videos: Using an image of a person and a text prompt, the model can generate a video with state-of-the-art results on character preservation and natural movement in video.
We’re continuing to work closely with creative professionals from across the field to integrate their feedback as we work towards a potential release. We look forward to sharing more on this work and the creative possibilities it will enable in the future.
When assigned a task on JIRA, an AI Worker will soon be able to finish the task assigned to it just as a human employee would do. It will create a roadmap, read the company docs, implement code, send PRs, communicate with others over Slack, join Zoom +
2
1
377
I believe by the end of 2026, these remote AI workers will be much more capable, and then it would make more sense for some companies to hire them instead of humans for many roles. What do you think?
1
1
226