When you write the prompt for any LLM, “why does your programming use the pronoun ‘I’ when everyone knows humans write the code providing these responses” and the services all respond, “I’m sorry if I’ve upset you” remember all that follows are hallucinations.
LLMS
-
Not All AI Research Focuses on Large Language Models
By
–
I wouldn’t assume that all of the research is going to LLMs.
-
Why LLMs Fail at ARC Tasks Despite Excelling on Exams
By
–
This may seem surprising to many people — if LLMs can score above human level on all those hard human exam benchmarks, why couldn't they do something as simple as ARC tasks, most of which seem trivially obvious to humans? It's because these tasks are mostly *new* — you won't
-
Formal Definition of Intelligence and Benchmark for LLMs
By
–
Here's a paper you can read about it. https://
arxiv.org/abs/1911.01547 It introduced a formal definition of intelligence, as well as benchmark to capture that definition in practical terms. Although it was developed before the rise of LLMs, current state-of-the-art LLMs such as Gemini -
Agent Lumos: Unified Modular Training for Open-Source Language Agents
By
–
Agent Lumos: Unified and Modular Training for Open-Source Language Agents Yin et al.: https://
arxiv.org/abs/2311.05657 #ArtificialIntelligence #DeepLearning #MachineLearning -
LLM Development Timeline: Research Before Deployment Economics
By
–
R typically comes before D. We have better insight in LLM timing than in most industries. It is over a year to go from startup to GPT-4 class model even with infinite money (Musk, Google). I don’t think that this tells us anything about whether we are in a bubble or a boom.
-
E5-Large-V2 and Nomic 1.5 embedding models exploration
By
–
I used E5-Large-V2 last year https://
til.simonwillison.net/llms/embed-par
agraphs
… Currently exploring @nomic 1.5: https://
github.com/simonw/llm-nom
ic-api-embed
… -
Closed API LLMs Still Lead Open Source Alternatives
By
–
For LLMs the closed API-only models are for the most part still so far ahead of the openly licensed alternatives that I'm sticking with them despite this Embedding models are different though: plenty of high quality models offer both an API and open weights I can run myself
-
API Model Reliability and Self-Hosted AI Systems Concerns
By
–
Given how hard it is to iterate on, evaluate and build dependable systems on top of these models the idea that a model you've tuned an application for can have its API turned off at any moment is pretty nerve-wracking So I lean towards API models I can run myself if necessary
-
Open Weights Model Reaches High Arena Ranking Despite License Constraints
By
–
Nice to see an open weights model showing up this high on the Arena leaderboard Worth noting that Command R is licensed Creative Commons Non-Commercial, so it's great for research but not a dependable backup against API shutdown for commercial purposes