OpenAI's Strawberry, LM self-talk, inference scaling laws, and spending more on inference

Q: How long is this episode of Interconnects?

This episode is 10 minutes long.

Q: When was this Interconnects episode published?

This episode was published on September 5, 2024.

Q: Can I download this Interconnects episode?

Yes, you can download this episode by clicking the download button on the episode player, or subscribe to the podcast in your preferred podcast app for automatic downloads.

from Interconnects · host Nathan Lambert

Whether or not scaling works, we should spend more on inference.This is AI generated audio with Python and 11Labs.Source code: https://github.com/natolambert/interconnects-toolsOriginal post: https://www.interconnects.ai/p/openai-strawberry-and-inference-scaling-laws00:00 OpenAI's Strawberry, LM self-talk, inference scaling laws, and spending more on inference01:51 OpenAI's Strawberry04:16 Self-talk in language models07:45 Inference scaling lawsFig 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/strawberry/img_006.pngFig 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/strawberry/img_021.pngFig 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/strawberry/img_023.pngFig 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/strawberry/img_037.png This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.interconnects.ai/subscribe

NOW PLAYING

OpenAI's Strawberry, LM self-talk, inference scaling laws, and spending more on inference

0:00 10:40

1×

No transcript for this episode yet

We transcribe on demand. Request one and we'll notify you when it's ready — usually under 10 minutes.

Share this episode

Similar Episodes

Top 5 Concerns Project Managers Have About Printed Electronics – And How to Solve Them

Jun 9, 2026 ·5m

Printing instead of a PCB – how to simplify IoT device design

May 20, 2026 ·8m

KNOWLEDGE THAT STAYS – A Mini Guide for Project Managers

May 12, 2026 ·4m

How printed electronics is changing the rules in IoT ?

Apr 28, 2026 ·7m

Ethics and sustainability in the purchasing process

Apr 22, 2026 ·8m

Tools for printed electronics designers – overview and recommendations

Apr 16, 2026 ·6m

Similar Podcasts

Hardware-Conscious Data Processing (ST 2023) - tele-TASK Prof. Dr. Tilmann Rabl Hardware development continuously advances, with different technologies improving at different pace. While the amount of transistors in a CPU package are growing, the single core performance is stagnating due to physical limitations. These trends require changes in data processing to keep database management systems efficient. In this lecture, we will take a look at current computer architectures and accelerator technologies and how they can be used for efficient data processing. We will cover CPU and memory architecture; the storage hierarchy; modern memory technolgoies, such as NVM and NVMe; fast interconnects, such as Infiniband, RDMA, and NVLink; and accelerators, such as GPUs and FPGAs. The course has a significant practical part, where the students learn to implement data structures and algorithms tailored to hardware concious data processing. Audistorium Stygian Catalyst Audistorium is a multi-genre spanning dark anthology audio drama created by Landon 'Lemon' Whisnant. From dread horror to absurdist comedy, Audistorium weaves a web of its own that interconnects It's stories in its own macabre, sometimes goofy way.Produced by Stygian Catalyst and co-creator of the Questionable Guide to Life Podcast.At the caring chiding of those close to us, we have decided to open up a way for people to contribute to the shows production, for the price of a simple cup of coffee, you can support Audistorium by clicking here for our Ko-Fi page.For contact, email us at [email protected],We can be found @AudistoriumPod on TwitterYou can find Landon <a href="https://open.acast.com/shows/653838418299010011ba94bc/episodes/@https://twitter.com/Lemjam Musical Tourism Synapset Synapset is a blitz collective formed in Barcelona, over a week in the beginning of April 2010 by Synapskollaps and reSet Sakrecoer. This album is based on experimenting with the risk of taking opportunities in life and reproduce them with machines. It questions the space existing between people and how music interconnects them. This album was written, recorded, mixed and mastered in 7 days.It's core formation is Synapskollaps and reSet Sakrecoer, with special appearance by Dr.Tikov and MC Charlot. Recorded In The FragleRock Studio v2.59, Barcelona. Cover photo by Patsy Boop, Edit by the Sakrecoer Design Robot. Mastered By Dr. Tikov9 tracks of pure kick and base!"Including amazing holiday pictures, healthy Sub-Vibes and pure feelings." - Basspistol.com"Congratulation on the release" - Goodkarma.ru The Undisputed Truth. Lily Stinson The undisputed truth…is within you.We’ll be diving into resonance beyond words. The truth we’re all searching for——LOVE. Simple. Direct. Digestible truth❤️ I’m not here to dull myself down and neither are you! A peak into limitless creation—- hosted by Lily (love)! I will reflect the truth within you——what interconnects and intertwines us all. Love. The simple truth humanity has forgotten about—-the cure of it all. The lion sleeps no more.

Frequently Asked Questions

How long is this episode of Interconnects?

This episode is 10 minutes long.

When was this Interconnects episode published?

This episode was published on September 5, 2024.

What is this episode about?

Whether or not scaling works, we should spend more on inference.This is AI generated audio with Python and 11Labs.Source code: https://github.com/natolambert/interconnects-toolsOriginal post:...

Can I download this Interconnects episode?