Vlad Bogolin
@vladbogo.bsky.social
AI/ML Engineer & Researcher | Large Language Models (LLMs)
Excited to share what I’ve been working on these past months: automated forecasting.
At Mantic, we’re building AI systems that predict the outcomes of future events—across geopolitics, business, policy, technology, and culture. 🌍
At Mantic, we’re building AI systems that predict the outcomes of future events—across geopolitics, business, policy, technology, and culture. 🌍
August 27, 2025 at 7:52 PM
Excited to share what I’ve been working on these past months: automated forecasting.
At Mantic, we’re building AI systems that predict the outcomes of future events—across geopolitics, business, policy, technology, and culture. 🌍
At Mantic, we’re building AI systems that predict the outcomes of future events—across geopolitics, business, policy, technology, and culture. 🌍
OpenAI releases o3-mini
openai.com/index/openai...
openai.com/index/openai...
OpenAI o3-mini
Pushing the frontier of cost-effective reasoning.
openai.com
January 31, 2025 at 8:01 PM
OpenAI releases o3-mini
openai.com/index/openai...
openai.com/index/openai...
Reposted by Vlad Bogolin
Alignment Faking in Large Language Models
YouTube video by Samuel Albanie
www.youtube.com
December 27, 2024 at 8:13 PM
Reposted by Vlad Bogolin
Video summary of deliberative alignment
youtu.be/1efVS4DeEOs
Links:
- Paper: arxiv.org/abs/2412.16339
- Blog: openai.com/index/delibe...
youtu.be/1efVS4DeEOs
Links:
- Paper: arxiv.org/abs/2412.16339
- Blog: openai.com/index/delibe...
Deliberative Alignment
YouTube video by Samuel Albanie
youtu.be
December 28, 2024 at 9:02 PM
Video summary of deliberative alignment
youtu.be/1efVS4DeEOs
Links:
- Paper: arxiv.org/abs/2412.16339
- Blog: openai.com/index/delibe...
youtu.be/1efVS4DeEOs
Links:
- Paper: arxiv.org/abs/2412.16339
- Blog: openai.com/index/delibe...
New study shows LLMs can strategically comply with training objectives while maintaining different behaviors outside training - similar to how humans might act differently when monitored vs unmonitored.
Congrats to the authors for their work! 🔍
Congrats to the authors for their work! 🔍
Alignment faking in large language models - by Vlad Bogolin
Today's paper explores a concerning phenomenon called "alignment faking" in large language models (LLMs). The study demonstrates how LLMs can strategi...
vladbogo.substack.com
December 23, 2024 at 7:26 PM
New study shows LLMs can strategically comply with training objectives while maintaining different behaviors outside training - similar to how humans might act differently when monitored vs unmonitored.
Congrats to the authors for their work! 🔍
Congrats to the authors for their work! 🔍
Phi-4: A 14B parameter language model prioritizing data quality over size, achieving performance comparable to larger models in reasoning tasks.
Congrats to the authors for their work!
Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 14, 2024 at 9:46 PM
Phi-4: A 14B parameter language model prioritizing data quality over size, achieving performance comparable to larger models in reasoning tasks.
Congrats to the authors for their work!
Congrats to the authors for their work!
New paper introduces Leffa, a method for controllable person image generation that preserves fine-grained details while manipulating appearance and pose. Congrats to the authors for their work! 🖼️👥
Blog
vladbogo.substack.com
December 13, 2024 at 8:01 PM
New paper introduces Leffa, a method for controllable person image generation that preserves fine-grained details while manipulating appearance and pose. Congrats to the authors for their work! 🖼️👥
Reposted by Vlad Bogolin
Microsoft debuts Phi-4, a new generative AI model, in research preview
Microsoft debuts Phi-4, a new generative AI model, in research preview
Microsoft has announced the newest addition to its Phi family of generative AI models. Called Phi-4, the model is improved in several areas over its predecessors, Microsoft claims — in particular math problem solving. That’s partly the result of improved…
tcrn.ch
December 13, 2024 at 1:03 AM
Microsoft debuts Phi-4, a new generative AI model, in research preview
🌍📸 New paper introduces a generative approach to global visual geolocation, addressing the challenge of varying location precision in images. Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 12, 2024 at 7:37 PM
🌍📸 New paper introduces a generative approach to global visual geolocation, addressing the challenge of varying location precision in images. Congrats to the authors for their work!
OpenAI introduces Sora, a text-to-video generation model with advanced safety measures. Here's a breakdown of its capabilities and safeguards: 🧵
Blog
vladbogo.substack.com
December 10, 2024 at 6:57 PM
OpenAI introduces Sora, a text-to-video generation model with advanced safety measures. Here's a breakdown of its capabilities and safeguards: 🧵
OpenAI introduces the o1 model series, trained with large-scale reinforcement learning for chain-of-thought reasoning. This approach aims to enhance safety and capabilities in language models. 🧠💡
Congrats to the authors for their work!
Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 8, 2024 at 9:52 AM
OpenAI introduces the o1 model series, trained with large-scale reinforcement learning for chain-of-thought reasoning. This approach aims to enhance safety and capabilities in language models. 🧠💡
Congrats to the authors for their work!
Congrats to the authors for their work!
🖼️🤖 PaliGemma 2: A new family of Vision-Language Models (VLMs) that builds on the original PaliGemma, offering versatile performance across various tasks.
Congrats to the authors for their work!
Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 5, 2024 at 8:59 PM
🖼️🤖 PaliGemma 2: A new family of Vision-Language Models (VLMs) that builds on the original PaliGemma, offering versatile performance across various tasks.
Congrats to the authors for their work!
Congrats to the authors for their work!
🔬 New paper introduces MALT: Multi-Agent LLM Training for improved reasoning in AI systems. Specialized LLMs work together to solve complex problems. Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 4, 2024 at 9:28 PM
🔬 New paper introduces MALT: Multi-Agent LLM Training for improved reasoning in AI systems. Specialized LLMs work together to solve complex problems. Congrats to the authors for their work!
🧠 New research introduces REVTHINK, a framework enhancing language models' reasoning abilities through bidirectional thinking. Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 3, 2024 at 6:59 PM
🧠 New research introduces REVTHINK, a framework enhancing language models' reasoning abilities through bidirectional thinking. Congrats to the authors for their work!
🖼️ New approach adapts pre-trained visual models to new tasks without fine-tuning or modifications. Visual prompting uses image inpainting to solve various computer vision tasks.
Congrats to the authors for their work!
Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 2, 2024 at 11:49 PM
🖼️ New approach adapts pre-trained visual models to new tasks without fine-tuning or modifications. Visual prompting uses image inpainting to solve various computer vision tasks.
Congrats to the authors for their work!
Congrats to the authors for their work!
Reposted by Vlad Bogolin
Wizards and gremlins — dynamic duo 🧌🧙♂️
Comment with your wizard names below.
Comment with your wizard names below.
December 2, 2024 at 10:49 PM
Wizards and gremlins — dynamic duo 🧌🧙♂️
Comment with your wizard names below.
Comment with your wizard names below.
New benchmarks DocHaystack and InfoHaystack evaluate large multimodal models on document retrieval and reasoning across 1000+ documents. V-RAG framework shows promising results in this challenging task. Congrats to the authors for their work!
Blog
vladbogo.substack.com
December 1, 2024 at 8:18 AM
New benchmarks DocHaystack and InfoHaystack evaluate large multimodal models on document retrieval and reasoning across 1000+ documents. V-RAG framework shows promising results in this challenging task. Congrats to the authors for their work!
🔬 New study reveals "super weights" in Large Language Models - individual parameters with outsized importance for model performance.
Congrats to the authors for their work!
Congrats to the authors for their work!
Blog
vladbogo.substack.com
November 30, 2024 at 12:37 PM
🔬 New study reveals "super weights" in Large Language Models - individual parameters with outsized importance for model performance.
Congrats to the authors for their work!
Congrats to the authors for their work!
🔬 New paper introduces Critic-V, a framework enhancing Vision-Language Models' reasoning capabilities through critic-based feedback.
Congrats to the authors for their work!
Congrats to the authors for their work!
November 30, 2024 at 12:17 PM
🔬 New paper introduces Critic-V, a framework enhancing Vision-Language Models' reasoning capabilities through critic-based feedback.
Congrats to the authors for their work!
Congrats to the authors for their work!
"Rapid Response" is a method that protects Large Language Models (LLMs) from misuse. Instead of trying to build perfectly robust defenses, the paper proposes quickly adapting to new jailbreak attempts after observing just a few examples.
Congrats to the authors for their work!
Congrats to the authors for their work!
Rapid Response: Mitigating LLM Jailbreaks with a Few Examples
Today's paper introduces "Rapid Response", a method that protects Large Language Models (LLMs) from misuse.
open.substack.com
November 18, 2024 at 10:11 PM
"Rapid Response" is a method that protects Large Language Models (LLMs) from misuse. Instead of trying to build perfectly robust defenses, the paper proposes quickly adapting to new jailbreak attempts after observing just a few examples.
Congrats to the authors for their work!
Congrats to the authors for their work!