gpt-5 is not needed to 100x the potential these models have we could stop all language model development today and we still haven’t scratched the surface of their capabilities here are a few non-obvious ways language models can be improved without creating any new models:
AI
-
Parallel Language Models Orchestrated by Conductor LLM
By
–
running language models in parallel with each one focused on a sub-task, all orchestrated by a conductor language model picture something like a massive tree of GPT models working on answering a single complex prompt
-
Language Models with Dynamic Memory and Reflection Capabilities
By
–
language models equipped with dynamic memory and the ability to reflect (e.g. Reflexion paper)
-
Local Language Models Using Cloud Models as Plugins
By
–
language models running locally on your machine that treat more complex language models in the cloud like plug-ins
-
Using Language Models to Pre-process and Optimize Your Prompts
By
–
using another language model for pre-processing your prompts to transform them into something that gives better responses if you want a prompt that you can use to already do this, read the prompt tip section here: https://
thepromptreport.com/p/report-5-eve
ryone-write-jailbreaks
… -
Building Belonging in Computer Science Communities
By
–
There are many ways to support fellow computer scientists beyond the ways you mention. One of the most important ones is to make people feel like they belong. I can imagine it's easy to feel like an outsider when few people in your field look like you.
-
Talent matters more than formal education in tech
By
–
Pour moi c'est : t'as des gens doué, et des gens pas doués. Qu'ils aient fait des études ou pas, et y'en a pleins qui apprennent sur le tas qui sont encore plus doués. Y'en a qui ont fait des etudes et qui sont pas doués. c'est histoire de personne et de competence pour le bon
-
AI Hallucination: Will Smith Eating Spaghetti Image Generation
By
–
An AI program hallucinating Will Smith eating spaghetti. Welcome to the future.
-
Hectare: Type-Directed Program Synthesizer 100x Faster
By
–
Want to get a program from just its type? Meet new type-directed synthesizer Hectare. Over 100x faster than state of the art on some problems — all with 10x less code. All thanks to a new kind of solver, ECTAs, solving a decades-old problem in synthesis: https://
bit.ly/3z2D5Ly -
The New Era of AI: Amazing and Incredibly Weird
By
–
The new era of AI is amazing, and incredibly weird. https://t.co/GsFYXDbNiA
— Will Knight (@willknight) 28 mars 2023The new era of AI is amazing, and incredibly weird.