Deep Learning for FUN @aneesha@aus.social
Brisbane
Joined March 2007
- Tweets56.7K
- Following2.3K
- Followers1.8K
- Likes64.1K
Aneesha Bakharia retweeted
here are some neuron hunters i'm tracking, who am i missing? @_lyraaaa_ @treetowntree @Sauers_ @jfonsecarivera @dorkitude @viemccoy @doyc_1 @neuronpedia
Aneesha Bakharia retweeted
Very excited to share Genie 4, our incredible interactive world model, bringing together real-time physics, reasoning, and audio! Simulations within simulations, makes you wonder just how deep the rabbit hole goes... I have a feeling we will be finding out soon enough...
Aneesha Bakharia retweeted
Recommended read. We are just starting to understand how AI shapes decisions and judgment.
At first glance, this might look like a simple, niche passion research project. However, it's an important research direction for understanding where AI's true value emerges and how to optimize agent-human collaboration.
If we can better understand when to defer to AI and when expert/human judgment matters, I think this unlocks tremendous value. Yes, it matters for organizations adopting AI, but I see it as an interesting way to better measure human-agent collaboration and discover new, optimal ways to augment collective intelligence.
Aneesha Bakharia retweeted
Remember a few weeks ago when I said people would simply Vibecode their own games and you guys didnβt believe me?
Well Google made a whole game engine you can prompt while playing
Gaps between player and creator is gone.
If Google makes this multiplayer, that kills steam, unity, unreal, PlayStation, Xbox, Nintendo, & Roblox all in one.
I genuinely donβt think the games industry will survive in its current form. Will undergo dramatic restructuring, similar to how newspapers & journalism has changed due to the internet and social media. Games wonβt go away. Creativity wonβt go away. But games industry is toast. Influencers & MMO community based vibecoding game building is about to eat it.
Had early access to Genie 4 from @GoogleDeepMind.
This is a fully generated, interactive world from a single starting prompt (environment + character) π€―
Feels like a massive step forward - you can now prompt to change the world and interact with it. With native audio!
Aneesha Bakharia retweeted
86% chance today of stratospheric wind reversal
Tuesday's Polar Vortex update ... trending strongly to unprecedented early warming event with reversal.
Model forecasts + climatology
polarvortex.earth/
Arctic Polar Vortex could break down as the 10 hPa wind at 60Β°N may reverse to easterly into early November.
The AIFS Ensemble shows 6% chance (3 members out of 50) while full physics EPS at 18/50 = 36%
Weather + GIS:
Polar Vortex Earth --> polarvortex.earth/
Aneesha Bakharia retweeted
People are building some seriously cool things with EmbeddingGemma 2 that itβs becoming hard to keep up with!
The community is taking its multimodal capabilities and running with them, building everything from in-browser WebGPU photo search to native Apple Silicon acceleration.
So, hereβs a round up of some amazing projects π§΅π:
Aneesha Bakharia retweeted
So excited to get this out:
LFM3 is finally here!
The first release in the LFM3 family contains six models in 2B-class size:
β’ LFM3-1.7B: to run thousands of tiny agents in parallel
β’ LFM3-VL-2B: now with video support
β’ LFM3-Audio-2B: Speech in AND speech out
β’ LFM3-Omni-2B: our first ever omni model
β’ d1-2B and d1-Omni-2B: with new types
They are all built on the same shared LFM3-1.7B-Base.
These are super powerful to fine-tune them into tiny specialists in your domain.
Weights on Hugging Face: huggingface.co/collections/Lβ¦
Introducing π»π΅πΌπΉ: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> π»π΅πΌπΉ-π·.π½π±: text, 256k context window.
> π»π΅πΌπΉ-π
π»-πΈπ±: vision-language model with video-input support for spatial understanding
> π»π΅πΌπΉ-π°ππππ-πΈπ±: tool calls, voice chat, reasoning in 18 languages.
> π»π΅πΌπΉ-πΎπππ-πΈπ±: an omni model combines vision, audio, and language.
> ππ·-πΈπ± and ππ·-πΎπππ-πΈπ±: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> π»π΅πΌπΉ-π
π»-πΈπ± is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> π»π΅πΌπΉ-π·.π½π± is #1 on the Pareto frontier of device intelligence πππππππ.ππππππ.ππ
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ π§΅
Aneesha Bakharia retweeted
NEW: Liquid AI just released LFM3 on Hugging Face. The model family brings text, vision, audio, omni, and decision models together on a new shared backbone (~2B params).
It can even do real-time video captioning locally in your browser on WebGPU with π€ Transformers.js!
Aneesha Bakharia retweeted
You don't see releases like this every day.
Liquid AI just released LFM3, a family of ~2B open-weight models for text, vision, audio, omni, and decision tasks, all on one shared backbone.
LFM3-1.7B is trained to run in agent harnesses like Hermes Agent and Pi, with a 256K context.
On WebArena-Verified, it scores 38.4 vs. 28.9 for Qwen3.5-9B in Liquid's evaluation. Models can be fine-tuned on a single GPU.
This series of models has huge potential for always-on local agents.
π€ Paid partnership
Introducing π»π΅πΌπΉ: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> π»π΅πΌπΉ-π·.π½π±: text, 256k context window.
> π»π΅πΌπΉ-π
π»-πΈπ±: vision-language model with video-input support for spatial understanding
> π»π΅πΌπΉ-π°ππππ-πΈπ±: tool calls, voice chat, reasoning in 18 languages.
> π»π΅πΌπΉ-πΎπππ-πΈπ±: an omni model combines vision, audio, and language.
> ππ·-πΈπ± and ππ·-πΎπππ-πΈπ±: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> π»π΅πΌπΉ-π
π»-πΈπ± is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> π»π΅πΌπΉ-π·.π½π± is #1 on the Pareto frontier of device intelligence πππππππ.ππππππ.ππ
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ π§΅
Aneesha Bakharia retweeted
This is really impressive and an area where Google continues to shine.
Introducing Genie 4 from @GoogleDeepMind, our most realistic and interactive world model yet π
Now you can use natural language to dynamically steer the world and its soundscape in real-time, whether you're changing the weather, moving the camera, or introducing new elements to the environment.
So, how does the model do that?
Behind the scenes, Gemini acts as the "brain" of the simulation. It translates your requests into actions, plans out complex events, watches the scene unfold to keep track of what's happening, and adjusts as you change the story. Genie 4 also introduces ambient audio and real-time interactive human dialogue that adapts on the fly.
Aneesha Bakharia retweeted
1/ You can now run a research swarm entirely on your Mac, model included.
Eight agents share one local 40B Flint model on a 48 GB M4 Pro: writing code, testing ideas and building on whatever works.
Give them a problem and a score to beat. Here's how π§΅
Aneesha Bakharia retweeted
I would definitely be for relatively more exact algorithms of the type of the trigonometric algorithm in grokking modular addition in toy model paper.
Or of the type of the line breaking algorithm.
Where are they?
When Models Manipulate Manifolds: The Geometry of a Counting Task
Aneesha Bakharia retweeted
Seedance 2.5 is UNLIMITED on Renoise this Halloween π
7 days of unlimited generations. No tricks, just treats.
Follow @renoiseai + RT. 2 winners get $200 in credits each (72H ONLY)
Aneesha Bakharia retweeted
This talk is a beautiful articulation of one of the most important design problems of our time. A few thoughts on planning with agents:
- I'm finding one of the best inputs to plans are recordings of live conversations among a team. this tends to capture a *ton* of peripheral useful context, that's hard to capture even in a solo voice recording -- because people elicit intent from each other. A generalization of this principle is that agents thrive on broad, high level, "messy" context - just like senior employees.
- I think visual outputs from agents are critical to complement text, but I'm actually of mixed minds on interaction. Yes, it's clearly nice for certain kinds of things like twiddling visual parameters. But most of the interactivity my agents produce feels mostly designed for dopamine reasons and not providing much utility. I also think interaction can inhibit deep thought compared to linear reading. I actually think for many deeper problems I would prefer top notch prose -- Richard Dawkins / Steven Pinker / Annie Dillard level -- from my agents over any kind of flashier output.
- One mental model I find consistently helpful for planning: the job is to establish a *design space* together between human and agent -- that is, a set of relevant dimensions of variation -- and then to efficiently explore the space. Getting the dimensions of the space right is actually one of the hardest parts. Once you have that, you can ask an agent to plant a bunch of starting points spread throughout the space, and then locally hillclimb from the best one.
New talk! About the divided human & agent worlds, boundary objects, and building thicker interfaces
I've been exploring how to improve "planning", aka. deciding what agents should build, how you want it built, & making the whole team part of that process
maggieappleton.com/planning-β¦
Aneesha Bakharia retweeted
The ospaic.org site is unreasonably cool.
Announcing the Call for Papers for Oxford Studies in Philosophy of AI and Computing!
Rapid AI progress is raising *so many* philosophical questions; but philosophers aren't as involved in the conversation as they should be, for many different reasons.
One is that, aside from some work in logic and philosophy of mind, philosophy of AI and computing has often sat on the margins of the discipline. One reason for *that* is that so many philosophers working on AI have had to do so basically on their own, figuring things out for themselves as they go along, without a real community around them.
OSPAIC is going to help fix this! We're going to have an annual conference and volume. The conference (in DC! where much of the future of AI will be decided) is especially important as we think about the nature of intellectual life in a world where AI systems can one-shot a superficially-plausible philosophy paper. OSPAIC sees philosophical progress as being about more than just the accumulation of true claims. The conference is a big part of our vision for how we contribute to philosophical progress.
And our aim is also to help philosophers be part of the broader public conversation! To that end we're also going to be launching, soon, a subcode on arXiv for philosophical foundations of AI (cs.pfai), and encouraging everyone who submits to OSPAIC to also upload their draft to arXiv (at present, use keyword PFAI).
We're also partnering with @cosmos_inst, who will be sponsoring an annual prize for the best paper on philosophy and AI, as well as helping fund the first round of the conference and Open Access publication for the volume. And with Cosmos we'll be highlighting some of the most interesting submissions to cs.pfai as they come in.
I'm the executive editor; Jeff Howard (UCL) and Rachael Sterken (University of Hong Kong) are associate editors; Ned Howells-Whitaker (JHU and Pittsburgh) is the assistant editor. We're grateful for the support of a wonderful advisory editorial board, including brilliant philosophers and computer scientists from academia, industry, and civil society.
The deadline for submissions is 11.59pm Feb 14, 2027, anywhere on earth. All details are at ospaic.org.
Aneesha Bakharia retweeted
This took me back to childhood playing in my sandbox with Tonka trucks and plastic soldiers.
MOSS the 3D printed open source litter picking robot on the beach for the first time. Tracks are well sized for sand: dry sand low dunes, damp sand and the waterline, no problem.
The big dry dunes near the entrance are another story, the ground clearance is low and it got stuck twice. A quarter of the way down they get smaller and it passes without trouble. To take the big ones it would need a bigger robot or more ground clearance.
It does not drive straight yet, I correct all the time, same as on the roadside video. That one is on me: the two tracks are not tensioned the same. Two tensioners per track, one sets the idler angle and the other the tension, and I only set them by eye after mounting. The encoders agree: on the road both motors turned the same within 1 %, so the drift is between the track and the ground, not in the motors.
Battery: 12.1 V at switch-on, 11.8 V at rest back home. The 21700 cells are on their way.
The rover was designed in Fusion with Codex in just a couple of weeks. @OpenAI @OpenAIDevs
Aneesha Bakharia retweeted
Turning complicated hardware into simple, fun lessons | ESP32 edition
Just like my Arduino app, this one shows you what the ESP32 can do, one circuit at a time.
Made with Opus 5.5
learn-esp32.vercel.app
Hardware looks scary until you break it down into small steps. So I built an app that takes you through Arduino one circuit at a time.
Made with Opus 5.5
learn-arduino-red.vercel.app
Aneesha Bakharia retweeted
today with @Thom_Wolf, we introduce the alignment hub, a central place for misalignment incidents designed with three goals:
1. set a standard for minimal transparency that is easy for labs to follow
2. improve clarity by centralizing information and sourcing
3. track the progression of both the incidents and how they are reported
we believe there is a minimal level of transparency required to properly understand incidents and their impact. we want this to be as simple as possible for labs to disclose, and we noticed that even things that seem obvious like reporting the incident or detection date are not standard today. we believe that the public should know the answers to these basic questions: was there monitoring (CoT? network?), did it happen during training? eval (general? cyber? alignment?), deployment? and more.. we also think that labs should be more transparent than this, and we think that this minimal level of transparency should evolve as the field advances
having a central place also helps a lot to connect incidents and build understanding. a good example is the irregular incidents which are all linked and caused by a human error of allowing internet during a CTF eval. we also track progression over time, things like the % of incidents disclosed by labs vs third parties, the time it takes for labs to disclose, and a tentative transparency score.
this is a field moving very fast and we will do our best to make sure this hub evolves with the challenges ahead. this is still in "beta" (lots of vibecoded components even if we want this to be agent first!), but we want to publish it early so that people iterate on it. we welcome feedback and contributions, we see this as a community effort!
our hope is that this makes it easier for researchers, labs and the general public to report, learn, understand and analyze safety incidents
huggingface.co/spaces/witnesβ¦
