Charlie Hou (@hou_char) profile picture

Research Scientist @GoogleDeepmind. Past: Researcher @Meta, PhD @CarnegieMellon, BSE @Princeton Opinions are my own

Joined September 2023
Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)Photo by Charlie Hou (@hou_char)
Charlie Hou retweeted
Excited to share Gemini 4 Argon! I know many people have been curious about what we’ve been building. We are sharing as soon as possible starting with trusted defenders in our Fairwind Program. We developed Argon with frontier capabilities in coding, knowledge work and cyber security defense. We look forward to making it more widely available as we go through testing.
Replying to @sundarpichai
Importantly Argon has frontier safeguards and we are rolling it out responsibly - it’s with the US gov’t and going to a set of trusted cyber defenders through our Fairwind Program today. We’re going to make it available as soon as we can and as safely as we can. So hold tight, lots more coming, and you’re going to see us iterating rapidly. blog.google/innovation-and-a…
86
104
10
1,792
185,724
Charlie Hou retweeted
Today we're publishing our first research blog, Understanding a Law Firm through Study. We're sharing a glimpse of a future where agents are trained with native memory:
31
64
34
620
668,280
Charlie Hou retweeted
Replying to @JensenHuang @nvidia
880
2,750
1,106
30,038
10,530,516
I worked for Peking University for a decade when China was loudly touting it desire to get more Chinese back to China. Let me tell you the dirty secret: Almost none of the returning Chinese who I counted as friends said or indicated them returning to China had anything to do with geopolitics. A spouse. Simply returning back home. An aging parent. The vast majority of time people moved back to China had zero to do with global politics, US politics, and everything to do with just what they wanted to do in life. I'm very sorry to disappoint whatever narrative you have
kimi’s founder & ceo got his phd at cmu. why didn’t he stay in america?!
126
131
22
2,212
321,263
Charlie Hou retweeted
David is one of those rare people with a deep inner calm, quiet conviction, and persistence, even a kind of elegance.
Replying to @JinWeiyang18434
Really enjoyed our chat today; thanks for stopping by! WISE is an awesome benchmark 😁
3
1
17
4,172
Charlie Hou retweeted
We’d like to introduce our paper on long-context positional encoding, centered on a simple principle:
2
2
1
12
28,245
Charlie Hou retweeted
Our @initc3org team of over 20 authors spent the last 6+ months wading through a huge body of work to systematize what we know (and what we don't!) in the space of AI x crypto. A few main takeaways:
2
10
1
36
6,051
Charlie Hou retweeted
Almost all animals sleep. Why don’t LMs? Introducing our new work on language model sleep. tl;dr : A periodic, recurrent “sleep” phase allows LMs to digest their context and transfer it into their weights, improving recall and reasoning on challenging tasks.
38
109
20
1,345
126,669
Charlie Hou retweeted
Sharing a super simple, user-owned memory module we've been playing around: nanomem The basic idea is to treat memory as a pure intelligence problem: ingestion, structuring, and (selective) retrieval are all just LLM calls & agent loops on a on-device markdown file tree. Each file lists a set of facts w/ metadata (timestamp, confidence, source, etc.); no embeddings/RAG/training of any kind. For example: - `nanomem add <fact>` starts an agent loop to walk the tree, read relevant files, and edit. - `nanomem retrieve <query>` walks the tree and returns a single summary string (possibly assembled from many subtrees) related to the query. What’s nice about this approach is that the memory system is, by construction: 1. partitionable (human/agents can easily separate `hobbies/snowboard.md` from `tax/residency.md` for data minimization + relevance) 2. portable and user-owned (it’s just text files) 3. interpretable (you know exactly what’s written and you can manually edit) 4. forward-compatible (future models can read memory files just the same, and memory quality/speed improves as models get better) 5. modularized (you can optimize ingestion/retrieval/compaction prompts separately) Privacy & utility. I'm most excited about the ability to partition + selectively disclose memory at inference-time. Selective disclosure helps with both privacy (principle of least privilege & “need-to-know”) and utility (as too much context for a query can harm answer quality). Composability. An inference-time memory module means: (1) you can run such a module with confidential inference (LLMs on TEEs) for provable privacy, and (2) you can selectively disclose context over unlinkable inference of remote models (demo below). We built nanomem as part of the Open Anonymity project (openanoymity.ai), but it’s meant to be a standalone module for humans and agents (e.g., you can write a SKILL for using the CLI tool). Still polishing the rough edges! - GitHub (MIT): github.com/OpenAnonymity/nan… - Blog: openanonymity.ai/blog/nanome… - Beta implementation in chat client soon: chat.openanonymity.ai Work done with amazing project co-leads @amelia_kuang @cocozxu @erikchi !!
8
45
4
306
75,513
I’m excited today to share Grapevine, a system that makes it really easy for AI agents to search company knowledge across, Slack, Notion, codebase, and more. The first app we’re launching with it is a company GPT that works remarkably well, better than any alternative today.
14
24
8
97
54,805
I've been on the team building Grapevine since the beginning, and in the span of a few months our internal company GPT grew from a hacky experimental demo into my most-used work AI tool.
I’m excited today to share Grapevine, a system that makes it really easy for AI agents to search company knowledge across, Slack, Notion, codebase, and more. The first app we’re launching with it is a company GPT that works remarkably well, better than any alternative today.
1
3
9
1,004
Charlie Hou retweeted
like everyone else i am hopping on the blog post trend gene.ttic.edu/blog/incomplet…
29
4
180
19,403
Charlie Hou retweeted
There are many algorithms for constructing pre-training data mixtures—which one should we use? Turns out: many of them fall under one framework, have similar issues, and can be improved with a straightforward modification. Introducing Aioli! 🧄 1/9
2
53
5
187
28,304
With Chameleon we showed that early fusion mixed modal LLMs can deliver strong improvements over unimodal and late fusion alternatives, however with this paradigm shift how do we rethink our core model architecture to optimize for native multimodality and efficiency? We introduce MoMa (Mixture of Modality-aware experts), a study of adaptive architectures delivering up to 4x FLOP efficiency (cheaper and smaller), we conduct a detailed study of adaptivity on modality, width, and depth. Check out @VictoriaLinML’s thread to learn more about this work!
1/n Introducing MoMa 🖼, our new sparse early-fusion architecture for mixed-modal language modeling that significantly boosts pre-training efficiency 🚀 (arxiv.org/pdf/2407.21770). MoMa employs a mixture-of-expert (MoE) framework with modality-specific expert groups. Given any interleaved mixed-modal token sequences, each group exclusively processes tokens of the designated modality with conventional MoE routing. This is joint work with amazing co-first authors @AkshatS07, @ArmenAgha and collaborators @AIatMeta – Liang Luo, @sriniiyer88, @ml_perception, @gargighosh and @LukeZettlemoyer.
8
27
7,810