ai-scholar.tech/articles/large…
posted at 13:50:19
RT @fladdict: 1/3に reading… ゲーム開発に生成AI 新興、コスト3分の1に - 日本経済新聞 www.nikkei.com/article/DGKKZO…
posted at 13:04:31
RT @kazunori_279: Vertex AI SearchとLLMでRAGを構成、実業務で3週間 使ってみた結果を詳細に解説。現場でRAGがどう役立つかが具体的に示されててありがたし。幻覚の問題はなかったのかな。ちなみに現在はPaLM APIが日本語対応したのでOpenAIなしでも構成可能。 #gcpja dev.classmethod.jp/articles/impro…
posted at 13:04:11
RT @_akhaliq: CodePlan: Repository-level Coding using LLMs and Planning paper page: huggingface.co/papers/2309.12… Software engineering activities such as package migration, fixing errors reports from static analysis or testing, and adding type annotations or other specifications to a codebase,… twitter.com/i/web/status/1… pic.twitter.com/LAjcoEg1xo
posted at 11:40:21
RT @llama_index: We’re excited to release full native support for THREE @huggingface embedding models (s/o @LoganMarkewich): 🧱 Base @huggingface embeddings wrapper 🧑🏫 Instructor embeddings ⚡️ Optimum embeddings (ONNX format) Full thread below 🧵. Checkout the guide: gpt-index.readthedocs.io/en/latest/exam… pic.twitter.com/KgVqOT08jL
posted at 11:02:10
RT @jerryjliu0: A huge bottleneck in scaling LLM/RAG systems to hundreds/thousands of docs (or more) is embedding speed ⏳ Here’s a simple trick to 3-10x your sentence_transformers throughput ⚡️: use the @huggingface Optimum library. Full native support in @llama_index: gpt-index.readthedocs.io/en/latest/exam… pic.twitter.com/8stfL3oVNV
posted at 09:55:46
RT @Ruin_ami: MI系の研究室を出ているから分かるけど、①はほぼ100%無理。 だって、情報系の人は化学やバイオの知識を得なくても年収の高い業界に行けるし、そもそも化学に興味が無い人ばっかりだもん。難易度云々以前の問題。 だから②しかない。化学者は頑張って。 twitter.com/kokonatsu2214/…
posted at 09:50:16
RT @bindureddy: The Ongoing Case For Open Source LLMs Custom LLMs, long context, and efficient inference Some folks believe that training open-source LLMs is a losing battle and a complete waste of time. They argue that the gap between closed models like GPT-4 and open models like Llama will… twitter.com/i/web/status/1… pic.twitter.com/JV8UEd1YWD
posted at 09:48:09
RT @NeelNanda5: This paper's been doing the rounds, so I thought I'd give a mechanistic interpretability take on what's going on here! The core intuition is that "When you see 'A is', output B" is implemented as an asymmetric look-up table, with an entry for A->B. B->A would be a separate entry twitter.com/OwainEvans_UK/…
posted at 09:31:15
RT @NeelNanda5: Unsurprisingly, the model has no issue with reversing facts in context! Intuitively, when I remember a fact A is B, it's closer to a mix of retrieving it into my "context window" and then doing in-context learning, rather than pure memorised recall.
posted at 09:30:58
RT @NeelNanda5: A better analogy might be in-context learning, where LLMs CAN use "variables". The text "Tom Cruise is the son of Mary Lee Pfeiffer. Mary Lee Pfeiffer is the mother of..." has the algorithmic solution "Attend to the subject of sentence 1 (Tom Cruise), and copy to the output"
posted at 09:30:37
RT @NeelNanda5: LLMs are not human! Certain things are easy for us and not for them, and vice versa. My guess is that the key difference here is that when detecting/outputting specific tokens, the LLM has no notion of a variable that can take on arbitrary values - a direction has fixed meaning
posted at 09:30:26
RT @abacaj: If you are running any local LLMs I would suggest scaling up as much as you can for the obvious benefit of increased model capacity and in context capabilities. This means you should try running a 34B codellama model at 4bit instead of a 13B or 7B model at higher precision
posted at 08:47:33
RT @hennabuta17: 「大谷翔平はただの夢だったのか」と題した米老舗雑誌「ザ・ニューヨーカー」の記事。 www.newyorker.com/sports/sportin… あまりにも良い内容だったので、要約をシェア。… twitter.com/i/web/status/1…
posted at 08:34:23
RT @syoyo: LLM 向け日本語データセットクリーニングと構築, 品質スコアリング値で binning(buketize)し, 最終的に train に回せる状態のデータセットを作る Beauty shot の script がついにできたよ🥹👊 あとはこれを trainer に feed させて LLM pretrain するだけや😌👊(たぶん) github.com/lighttransport… pic.twitter.com/rjh5cQlU7P
posted at 08:31:10
RT @llama_index: We've added one of the first API specs for an LLM-powered agent to interact with a graph db (@neo4j) 🔥: 📣 Not just "text-to-cypher", use it as a tool with full agent reasoning (CoT, usage w/ other tools) Huge props to @ParienteShahaf for this: llamahub.ai/l/tools-neo4j_db pic.twitter.com/OLkCojgxJW
posted at 08:07:44
RT @shirayu: 岡野原さんの講演「生成モデルは世界をどのように理解しているのか」(2023/05)の19ページが参考になりそう hillbig.github.io/ISM_Symposium2… twitter.com/shirayu/status… pic.twitter.com/ffpWwLH0CU
posted at 08:01:12
RT @keio_smilab: 新学術領域「対話知能学」2023年度公開シンポジウムにおいて、「基盤モデルはロボティクスをどう変えるのか」という題で招待講演を行いました。 スライドを公開しました。 speakerdeck.com/keio_smilab/ho…
posted at 08:00:31