Expectation: I need more deep learning engineers to train better models
Reality: You need prompt engineers and LLM Ops (not sure what to call it (?), post-LLM above-API infra, langchain & friends)
– training is centralizing into megamodels – not fully played out yet but trending
LLMS
-
Deep Learning Skills Shift: From Training to Prompt Engineering
By
–
-
Model-Written Code Security Improves Across Generations
By
–
Security of model-written code increasing with each model generation:
-
Foundation Models Delivering Enterprise Value: Practical Implications
By
–
3 practical implications of using foundational models to get to enterprise value, via @SnorkelAI Foundation Model Summit 2023. Read the full transcript or watch the full replay at https://
snorkel.ai/sambanova-a-pr
actical-approach-to-delivering-enterprise-value-with-foundation-models/
… #ai #nlp #enterpriseai -
GPT-4 demonstrates zero grammar errors compared to humans
By
–
Contrairement à vous, #GPT4 fait ZÉRO faute d’orthographe et de grammaire Votre tweet comporte 18 fautes Il vous faut bosser pour rattraper #GPT4 https://
t.co/QwnVoT4vtO -
Comparing davinci and chat API models for task performance
By
–
I like davinci, he’s my homie (also I’ve heard mix reaction in 3.5 and 4 for these types of tasks). Would love to see others try w the chat api!
-
ChatGPT-4 vs Bard: Creative Task Showdown Analysis
By
–
We pitted two chatbots ChatGPT-4 and Bard against one another on tasks that humans would need creative thinking for. Here’s how they did, and what the FT’s experts thought of it –
-
LLM Survey Citations Expectation vs Reality in AI Research
By
–
Expectation: Writes LLM survey hoping to get tons of citations. Reality: No one writes LLM papers anymore
-
Politicians Ignoring AI Tsunami: Society Unprepared for GPT5
By
–
En 2050, nos petits-enfants seront stupéfaits : « 8 mois avant la sortie de GPT5, nos grands-parents débattaient sur une affaire de trottinettes à Paris » Les politiciens étaient TOTALEMENT IRRESPONSABLES Personne ne préparait la société au Tsunami technologique de l’IA
-
32k Context Length: Exploring AI Capabilities and Risks
By
–
If I know anything about us, once we have 32k context length… we’ll think of reasons we need more. But yes, definitely risks w this – tho still a worthy exploration as we better understand these tools.
-
LLM Compression Techniques and Practical Implementation
By
–
Oh nice! I took a couple stabs at “LLM compression” as I’d call this, but to no avail. Need to try this.