Explained here: https://
openai.com/research/effic
ient-training-of-language-models-to-fill-in-the-middle
…
@goodside
-
Research on Efficient Training of Language Models for Text Completion
By
–
-
Adding ‘Jar Jar Binks’ Style to AutoGPT Prompts for Mesa-Optimization
By
–
PR to add “in the style of jar jar binks” to all autogpt prompts so we get mesa-optimized
-
LLMs Immortalize Documented Knowledge and Diminish Traditional Expertise
By
–
i remember how fun the oreilly perl book was as a kid, to know the whole thing. our shared, slow-moving technical knowledge is becoming less important maybe. but LLMs make the parts we document well immortal. if the model knows, someone always will
-
LLMs Poised to Revolutionize Old CLI Tools and Libraries
By
–
a renaissance is coming for old CLI utils, plotting libraries, video rendering pipelines, big libraries with ugly APIs. LLMs are going to bring users to jupyter and matplotlib and cv2 and ffmpeg that have never seen a shell or written a line of python
-
Natbot: Defining Custom Languages with Task-Specific Opcodes for AI
By
–
Often the trick is to define your own simple language with task-specific opcodes — check out natbot for inspiration: https://
github.com/nat/natbot -
LLM Computational Limits and Novel Hash Collisions
By
–
Problems with calculation run deeper than just tokenization though — only so much computation can happen per token. E.g. an LLM could never predict a log of novel hash collisions found via brute force search.
-
AI Models and Retrieval Augmentations Expected in the Legal Sector
By
–
Haven’t seen much specific beyond the bar exam stuff in the news but I would assume models (or at least retrieval augmentations) for law are coming
-
Exploring the Fundamental Limits and Patching Mechanisms of LLMs
By
–
I feel like we need a unified theory of what LLMs can’t do. Most issues (calculation, string ops, grounding) can be patched (via tools, retrieval) so we see them as trivial, not as omens of subtler but similarly fundamental limits. But why is this true?
-
Discussion on Prompt Engineering, RLHF, AI Safety, and AGI
By
–
Got to chat about prompt engineering, RLHF, LLM red teaming, AI safety, and AGI with @labenz on the @CogRev_Podcast — thanks so much for having me, Nathan!
-
Customizing Embeddings and Comparing Distance Metrics in AI
By
–
Nice — curious how this compares to learning a modified distance metric as shown here: https://
github.com/openai/openai-
cookbook/blob/main/examples/Customizing_embeddings.ipynb
…