Hamza Merzić
@hamzamerzic.info
Staff Research Engineer @ Google DeepMind
PhD Candidate @ UCL
🇵🇸
PhD Candidate @ UCL
🇵🇸
EEML 2025 in Sarajevo 🇧🇦 is now open for applications until 31 March! 🚀
If you're a student, researcher, or industry professional passionate about AI, this is your chance to learn from top experts and connect with an amazing community.
Help spread the word! 🌍 Scholarships available—see below.
If you're a student, researcher, or industry professional passionate about AI, this is your chance to learn from top experts and connect with an amazing community.
Help spread the word! 🌍 Scholarships available—see below.
Applications are now open for EEML 2025 in Sarajevo, Bosnia and Herzegovina, 21-26 July! 🎉
Learn from top AI researchers and connect with peers in Sarajevo 🇧🇦, a historical crossroads of East and West. Needs-based scholarships are available.
Deadline: 31 March 2025.
Learn from top AI researchers and connect with peers in Sarajevo 🇧🇦, a historical crossroads of East and West. Needs-based scholarships are available.
Deadline: 31 March 2025.
February 7, 2025 at 3:36 PM
EEML 2025 in Sarajevo 🇧🇦 is now open for applications until 31 March! 🚀
If you're a student, researcher, or industry professional passionate about AI, this is your chance to learn from top experts and connect with an amazing community.
Help spread the word! 🌍 Scholarships available—see below.
If you're a student, researcher, or industry professional passionate about AI, this is your chance to learn from top experts and connect with an amazing community.
Help spread the word! 🌍 Scholarships available—see below.
The big selloff coming from DeepSeek results might be overlooking: 1) the compute cost of proving that large-scale RL is able to enhance LLM reasoning capabilities and 2) the increased inference-time compute demand of the thinking models paradigm.
January 27, 2025 at 4:11 PM
The big selloff coming from DeepSeek results might be overlooking: 1) the compute cost of proving that large-scale RL is able to enhance LLM reasoning capabilities and 2) the increased inference-time compute demand of the thinking models paradigm.
Love this line of work. Can't wait to try it out!
Introducing the smollest VLMs yet! 🤏
SmolVLM (256M & 500M) runs on <1GB GPU memory.
Fine-tune it on your laptop and run it on your toaster. 🚀
Even the 256M model outperforms our Idefics 80B (Aug '23).
How small can we go? 👀
SmolVLM (256M & 500M) runs on <1GB GPU memory.
Fine-tune it on your laptop and run it on your toaster. 🚀
Even the 256M model outperforms our Idefics 80B (Aug '23).
How small can we go? 👀
January 23, 2025 at 1:40 PM
Love this line of work. Can't wait to try it out!
I love intuitive explanations like this one, especially when they're about my favorite quantities like entropy! Share more if you have them!
Entropy is one of those formulas that many of us learn, swallow whole, and even use regularly without really understanding.
(E.g., where does that “log” come from? Are there other possible formulas?)
Yet there's an intuitive & almost inevitable way to arrive at this expression.
(E.g., where does that “log” come from? Are there other possible formulas?)
Yet there's an intuitive & almost inevitable way to arrive at this expression.
December 10, 2024 at 11:21 PM
I love intuitive explanations like this one, especially when they're about my favorite quantities like entropy! Share more if you have them!
It's surreal to witness the web of lies this world is built upon unraveling before our eyes...
December 5, 2024 at 3:54 PM
It's surreal to witness the web of lies this world is built upon unraveling before our eyes...
Reposted by Hamza Merzić
Introducing 🧞Genie 2 🧞 - our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds, playable for up to a minute. We believe Genie 2 could unlock the next wave of capabilities for embodied agents 🧠.
December 4, 2024 at 4:01 PM
Introducing 🧞Genie 2 🧞 - our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds, playable for up to a minute. We believe Genie 2 could unlock the next wave of capabilities for embodied agents 🧠.
Buying property in the UK is wild!
You might know about the whole leasehold/freehold stuff, but I just learned about chancel repair liability where you could be responsible for fixing a nearby church based on some medieval arrangements that are costly to trace.
So people just get insurance instead.
You might know about the whole leasehold/freehold stuff, but I just learned about chancel repair liability where you could be responsible for fixing a nearby church based on some medieval arrangements that are costly to trace.
So people just get insurance instead.
December 4, 2024 at 3:15 PM
Buying property in the UK is wild!
You might know about the whole leasehold/freehold stuff, but I just learned about chancel repair liability where you could be responsible for fixing a nearby church based on some medieval arrangements that are costly to trace.
So people just get insurance instead.
You might know about the whole leasehold/freehold stuff, but I just learned about chancel repair liability where you could be responsible for fixing a nearby church based on some medieval arrangements that are costly to trace.
So people just get insurance instead.
New work by my colleague @theo-brown.bsky.social
Hope to see more stuff like this on here!
Hope to see more stuff like this on here!
📣 If you've got an objective that exhibits symmetries, you should be using invariant kernel BO 📣
🚀 More sample efficient than constrained/naive BO!
🚀 More compute efficient than data augmentation!
🧵 1/4
#NeurIPS2024 #BayesianOptimisation #ai
🚀 More sample efficient than constrained/naive BO!
🚀 More compute efficient than data augmentation!
🧵 1/4
#NeurIPS2024 #BayesianOptimisation #ai
November 21, 2024 at 2:12 PM
New work by my colleague @theo-brown.bsky.social
Hope to see more stuff like this on here!
Hope to see more stuff like this on here!
I try hard not to overthink the first post on a new platform… but somehow always end up trying too hard.
November 21, 2024 at 1:47 PM
I try hard not to overthink the first post on a new platform… but somehow always end up trying too hard.