Sil Hamilton
@srhm_ca
Followers
666
Following
1K
Media
117
Statuses
394
I work with language models / grad student @Cornell, previously @McGill, talks @cs50, @ComputerHistory
Ithaca, NY
Joined December 2021
If you're interested in applying language models to cultural domains this is a great program - highly recommended!
Interested in working on AI and storytelling with .txtlab? We're offering a 2-year stipend for our MA in Digital Humanities program. Info attached. https://t.co/cCWQkJsIsL
1
1
2
Happy new year! I've written a retrospective on early AI research to finish off the year: https://t.co/NVyptB1S8Q — writing it brought me to https://t.co/ooL9ElndvO which has a treasure trove of early documents on AI research at MIT. Worth checking!
0
2
3
Tim Dettmers of LLM.int8() fame on quantizing MoE systems. Hoping this means Q2_K will be more competitive! It'd be a blessing for my 32GB system.
@srush_nlp @4evaBehindSOTA Regular transformers are notoriously difficult to sparsify, this is even true for the FFN layers in MoE transformers. But MoE layers are very different. You can also quantize them to 1 bit without any problem, but sparsification gives you better memory benefits than 1-bit quant.
0
0
1
AI needs journalism more than journalism needs AI: @ComputerHistory has published my article (co-written with @jenny8lee) on open-source #AI with other great pieces by Google VP @richardgingras, @BigLocalNews' Cheryl Phillips, and more! Read here: https://t.co/z3Y5pSiIYU
0
2
5
If you take a peek at the ngram viewer for 'think tank' there's a clear dip during the Great Recession... 🤔
1
0
2
Ecstatic to announce I'll be doing a second course with @JCourses on #GenAI for journalists! Learn how to build and run useful agents on your laptop with Jupyter notebooks and LangChain—it starts in just a week! https://t.co/x3zL7sgQ4f
0
0
6
And I’m on my way back to Canada! Lithuania is lovely, and I learned EU media might be less hostile to AI than I’d expect. Write-up forthcoming…
0
0
0
Here in Lithuania to give a talk on North American newsroom policies vis a vis AI—it'll be interesting to get the European perspective! Will report back with what I've learned...
0
1
6
Moreover—I'd like this to be used by data scientists not researching and developing language models! TensorBoard can deal with not-tensors, but it isn't necessarily designed to promote it. Something built into matplotlib's pyplot would be nice.
0
0
1
You do have Tensorflow's TensorBoard and Embedding Projector which are decent, but I'd like something I can easily call up from a Juptyer notebook... and that doesn't necessarily imply TF.
projector.tensorflow.org
Visualize high dimensional data.
1
0
1
Interesting how few solid RAG/embedding tools have hit the mainstream despite traction in the dev space—you still need to wrangle Python/the CLI to get anywhere!
1
0
2
DocPlot now supports .pdf and .docx files—I found it surprisingly simple to extract text from them. https://t.co/qDZKJeb0YI
Threw together a small website on Sunday: it lets you run semantic searches on documents embedded in the browser, privately. Try it out! https://t.co/Hp4U1L0nRd
2
0
2
Threw together a small website on Sunday: it lets you run semantic searches on documents embedded in the browser, privately. Try it out! https://t.co/Hp4U1L0nRd
0
0
7
Enjoying folks’ creative #journalism & #AI projects at #MediaParty 2023 in Buenos Aires. LLMs are malleable! Great for hackathons—so many untapped use cases.
0
1
7
Such a great two days here at the Kennedy Space Center. Thank you @CommHIT for hosting the conference—and a big thank you to @JaroneLeeMD & @HealthTechWB for inviting me to talk about language models.
0
3
10
I'm quoted in this month's @Wired! In @vauhinivara's great piece on AI co-writing, wherein she argues GPT-3.5 has "the witless efficiency of a stapler." I agree. For LLMs, words are concepts, and concepts are words—RLHF is linguistic relativity in action.
0
2
9
Super excited to announce I'll be keynoting this October at #MediaParty 2023 in Buenos Aires in my role as researcher-in-residence at @HacksHackers—looking forward to talking media & AI with people from across Latin America!
0
0
4