Ergebnisse
347 ArtikelSortieren
2:13:07EN
Cognitive Revolution "How AI Changes Everything"· 27. Juli 2023PodcastHalf an Hour to a Circuit in GPT-2 Small — and a Quadratic Wall Before GPT-3Moderation: Nathan LabenzArthur Conmy walks Nathan Labenz through ACDC, the algorithm that automates the most tedious step of mechanistic interpretability: finding which transformer components actually do a task.
4:59EN
Matthew Berman· 25. Juli 2023Testing Airoboros L2 70B on Coding, Writing and ReasoningMatthew Berman · Forward FutureMatthew Berman runs a small set of prompts against a Llama 2 fine-tune, finding useful structured output alongside failures on code and logic.
9:40ENMatthew Berman· 20. Juli 2023Testing Llama 2 70B on Coding, Writing and ReasoningMatthew Berman · Forward FutureMatthew Berman tries Meta’s 70-billion-parameter chat model through a hosted interface, highlighting both useful outputs and inconsistent reasoning.
5:53ENMatthew Berman· 18. Juli 2023Petals: Distributing Large Language Model Inference Across DevicesMatthew Berman · Forward FutureIn 2023, Matthew Berman explains Petals as a way to use large models through a network of participating computers rather than loading every layer on one device.
8:04ENMatthew Berman· 11. Juli 2023WizardLM 13B v1.1 Tested on Coding, Writing and LogicMatthew Berman · Forward FutureMatthew Berman tries an instruction-tuned model on a small set of prompts, with mixed results and several subjective scoring choices.
12:20ENMatthew Berman· 10. Juli 2023Testing UltraLM 13B: A Small Model’s Strengths and Reasoning ErrorsMatthew Berman · Forward FutureA 2023 local test finds useful writing and simple coding alongside failures in games, arithmetic and logic.
1:33:56ENCognitive Revolution "How AI Changes Everything"· 6. Juli 2023PodcastThe AI Scouting Report: 90 Minutes From GPT-4's Exam Scores to Whether Models UnderstandModeration: Nathan LabenzNathan Labenz spends a solo episode walking through what general-purpose models can do, how they are trained, how capabilities unlock, and whether they understand anything at all.
18:08ENMatthew Berman· 29. Juni 2023Six Language Models Compared on an Informal 2023 Prompt SetMatthew Berman · Forward FutureMatthew Berman tests coding, writing and reasoning with H2O GPT, finding a tie under a small and subjective rubric.
1:32:22ENCognitive Revolution "How AI Changes Everything"· 29. Juni 2023PodcastA Million Bytes and 256 Symbols: Meta's MEGABYTE Wants to Throw the Tokenizer AwayModeration: Nathan LabenzLili Yu of Meta AI explains a multiscale transformer that predicts million-byte sequences, skips tokenization, and aims to model any file as raw bytes.
1:52:46ENCognitive Revolution "How AI Changes Everything"· 27. Juni 2023PodcastCharge Neural Nets by Connection Length, and the Black Box Splits Into Modules You Can SeeModeration: Nathan LabenzMIT physicist Ziming Liu walks Nathan Labenz through BIMT, a brain-inspired training trick that sparsifies networks until you can read them by looking.
13:50ENMatthew Berman· 23. Juni 2023MPT-30B: Running MosaicML’s Model Locally and Testing Its Chat VariantMatthew Berman · Forward FutureMatthew Berman reviews MosaicML’s June 2023 release, demonstrates a local setup, and puts the hosted chat model through an informal prompt rubric.
4:36ENMatthew Berman· 20. Juni 2023Testing WizardCoder 15B as a Local Coding AssistantMatthew Berman · Forward FutureMatthew Berman runs the 2023 WizardCoder model locally, testing small Python exercises and a longer Snake game.
59:28ENCognitive Revolution "How AI Changes Everything"· 16. Juni 2023PodcastFrom $500,000 to $50,000: MosaicML Made Training Your Own Model a Priced ProductModeration: Nathan LabenzJonathan Frankle and Abhi Venigalla explain who is training custom LLMs, why extraction beats chat, and how ALiBi stretches a 65k context window.
2:00:19ENCognitive Revolution "How AI Changes Everything"· 6. Juni 2023Podcast1.5 Million GPT-4 Fairy Tales Let a 33-Million-Parameter Model Out-Reason GPT-2Moderation: Nathan LabenzTry this fill-in: Lily likes cats and dogs. She asked her mom for a dog. Mom said no, so she asked for a…
11:49ENMatthew Berman· 1. Juni 2023Testing Wizard-Vicuna 30B: Responses, Reasoning, and CodeMatthew Berman · Forward FutureMatthew Berman examines a model marketed as uncensored, demonstrates a 2023 cloud setup, and runs an informal collection of tests.
1:25:35ENCognitive Revolution "How AI Changes Everything"· 21. Mai 2023PodcastTwo and a Half Years From 33.3% to 86.5%: Google Sent Foundation Models to Medical SchoolModeration: Nathan LabenzVivek Natarajan walks Nathan Labenz through Med-PaLM and Med-PaLM 2: the data-light techniques, the nine-axis clinician evals, and why a high USMLE-style score is the wrong trophy.
42:18EN
No Priors: AI, Machine Learning, Tech, & Startups· 18. Mai 2023Podcast · 2 StimmenTen Percent of Search Is Health — and the Doctor Is Googling TooModeration: Sarah Guo, Elad GilGoogle researcher Karan Singhal on Med-PaLM 2, the real bar for medical language models, and why medicine is a harder alignment lab than chatbots.
11:07ENMatthew Berman· 16. Mai 2023Testing MosaicML MPT-7B Chat and StoryWriter in 2023Matthew Berman · Forward FutureMatthew Berman examines MPT-7B model variants, their licenses and a mixed set of local tests.
5:24ENMatthew Berman· 12. Mai 2023Google I/O 2023: Bard, PaLM 2 and AI Across Google's ProductsMatthew Berman · Forward FutureMatthew Berman reviews Google's 2023 AI announcements and weighs its focus on improving established products.
4:14:33EN
Lex Fridman· 9. Mai 2023Podcast · 2 StimmenThe Next Token Is Not a Proof: Stephen Wolfram on ChatGPT and Computational TruthGast: Stephen Wolfram · Moderation: Lex FridmanIn his fourth conversation with Lex Fridman, Stephen Wolfram contrasts large language models with Wolfram|Alpha, then follows computational irreducibility from ChatGPT through observers, thermodynamics, and the three great theories of twentieth-century physics.
9:58ENMatthew Berman· 26. Apr. 2023HuggingChat’s 2023 Debut: Testing Writing, Reasoning and Python CodeMatthew Berman · Forward FutureMatthew Berman tries the early OpenAssistant-powered HuggingChat on writing, arithmetic, logic and a Snake game, then compares a few outputs with GPT-3.5 and GPT-4. The results illustrate both promise and limitations in this small historical test.
5:47ENMatthew Berman· 19. Apr. 2023Sam Altman on Scaling AI Models: Matthew Berman Examines OpenAI’s 2023 DirectionMatthew Berman · Forward FutureA 2023 discussion of model scaling, possible ways to improve GPT-4 and the safety priorities OpenAI described at the time.
40:25ENNo Priors: AI, Machine Learning, Tech, & Startups· 6. Apr. 2023Podcast · 2 StimmenA 6-Billion-Parameter Clone of ChatGPT’s Core TrickModeration: Sarah Guo, Elad GilDatabricks CTO Matei Zaharia tells Sarah Guo and Elad Gil how Dolly learned instruction-following on fifty times less data than GPT-3 — and why he would rather look facts up than bake them into a trillion-parameter network.
6:24ENMatthew Berman· 6. Apr. 2023BloombergGPT: Mixing Financial and General Data for a Language ModelMatthew Berman · Forward FutureMatthew Berman examines the 2023 finance-focused research model, its training corpus, evaluation and decision not to release weights.Mehr laden24 von 347 angezeigt