Alle afleveringen van Interconnects – pagina 4 van 4

Why we disagree on what open-source AI should be

3 april 2024 · 9 min

Afspelen

Last minute title change from: The tech industry can't agree on what open-source AI means. That's the process.
How to read what multiple people mean by the word openness and see through the PR speak.
This is AI generated audio with Python and 11Labs.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/flavors-of-open-source-ai Lees meer

0:00 The tech industry can't agree on what open-source AI means. That's the process.
2:45 1. Effective Accelerationists, Techno-Optimists, capitalists, etc.
3:39 2. Scientists, promoting understanding and transparency
5:16 3. Inclusion, public interest, and fighting concentration of power
6:19 4. Freedom advocates
7:25 Dissecting "openness"

Fig 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/openness/img_004.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

DBRX: The new best open LLM and Databricks' ML strategy

29 maart 2024 · 17 min

Afspelen

Databricks' new model is surpassing the performance of Mixtral and Llama 2 while still being in a size category that's reasonably accessible.
This is AI generated audio with Python and 11Labs.
Source code: https://github.com/natolambert/interconnects-tools
https://www.interconnects.ai/p/databricks-dbrx-open-llm Lees meer

00:00 DBRX: The new best open model and Databricks' ML strategy
03:36 The DBRX narrative
07:33 Databricks' open LLM (and AI) strategy
09:42 Playing with DBRX Instruct
14:54 Digging for details

Fig 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_007.png
Fig 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_012.png
Fig 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_023.png
Fig 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_045.png
Fig 5: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_047.png
Fig 6: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_059.png
Fig 7: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_066.jpeg
Fig 8: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/dbrx/img_068.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

Evaluations: Trust, performance, and price (bonus, announcing RewardBench)

21 maart 2024 · 13 min

Afspelen

Evaluation is not only getting harder with modern LLMs, it's getting harder because it means something different.
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/evaluations-trust-performance-and-price Lees meer

00:00 Evaluations: Trust, performance, and price (bonus, announcing RewardBench)
03:14 The rising price of evaluation
05:40 Announcing RewardBench: The First reward model evaluation tool
08:37 Updates to RLHF evaluation tools

YouTube code intro: https://youtu.be/CAaHAfCqrBA

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/evals/img_026.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/evals/img_030.png
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/evals/img_034.png
Figure 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/evals/img_040.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

Model commoditization and product moats

13 maart 2024 · 11 min

Afspelen

Where moats are tested now that so many people have trained GPT4 class models. Claude 3, Gemini 1.5, Inflection 2.5, and Mistral Large are here to party.
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/gpt4-commoditization-and-moats Lees meer

00:00 Building LLM moats despite the commoditization of GPT4
04:38 The Open's opportunities
08:02 It's amazing people still think LLMs aren't going to be useful
09:50 Things that are coming

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/moats/img_004.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/moats/img_028.png
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/moats/img_032.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

The koan of an open-source LLM

6 maart 2024 · 23 min

Afspelen

A proposal for a new definition of an "open source" LLM and why no definition will ever just work.
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/an-open-source-llm Lees meer

00:00 The koan of an open-source LLM
03:22 A new naming scheme for open LLMs
07:09 Pivot points and politics
08:16 Claude 3, arms race, commoditization, and national security
10:01 Doomers debunking bio risks of LLMs themselves
11:21 Mistral's perceived reversal and the EU
13:22 Messy points: Transparency, safety, and copyright
13:32 The muddling of transparency
15:22 The muddling of "safety"
16:30 The muddling of licenses and copyright
20:12 Vibes points and next steps

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/open-source/img_046.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/open-source/img_064.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

Interviewing Louis Castricato of Synth Labs and Eleuther AI on RLHF, Gemini Drama, DPO, founding Carper AI, preference data, reward models, and everything in between

4 maart 2024 · 1 uur 26 min

Afspelen

This interview is available on podcast players and YouTube. Lees meer

I’m excited to bring you another interview! This one is a deep dive right in my wheelhouse — all things RLHF. Louis Castricato is probably the hidden star of RLHF in the open. I’m not sure anyone who can speak freely knows as much as him. As I’ve said again and again on Interconnects:

Giving a voice to researchers is the best way to cut through the noise and understand what is happening with core developments of LLM technologies.

Louis recently has been founding a new startup focused on synthetic data for alignment, Synth Labs, and is a researcher at Eleuether AI. This interview should speak for itself, and it’ll need re-listens, even for myself. The list of topics we cover touches on pretty much every major and minor issue facing model fine-tuning. Please reach out or comment if there’s a paper we mention that I didn’t link before. Happy to dig it up for you.

For more on Synth Labs, there was a profile in Bloomberg from Rachel Metz.

This post is very technical, more than usual. If you’re having a hard time with it, I suggest you listen to my RLHF 201 post on Latent Space first.

Chapters

These are generated with smol-podcaster with moderate edits.

High-level chapters

* 00:00:00: Introduction

* 00:01:24: Gemini News and RLHF’s Part in it

* 00:09:05: Long Context, In-Context, and Multimodal RLHF

* 00:21:20: What are people missing about RLHF these days?

* 00:30:30: OpenAI's Influence and the Need for Alternatives

* 00:39:20: Synth Labs and the Future of Alignment

* 00:55:00: Evaluation Talk p2: Open-ended Evaluation and Data Diversity

* 00:59:20: Algorithm Roundup: PPO, DPO, KTO, IPO

* 01:18:38: CarperAI, Early Days of RLHF, Reflecting on ChatGPT

Detailed chapters

* 00:00:00: Introduction and Overview of RLHF

* 00:02:02: Gemini News, Custom Demographics in Image Prompts, and Controllability Issues in AI Models

* 00:05:21: Fixing Biases in AI Models Post-Training, Representation in AI Dat…

How to cultivate a high-signal AI feed

28 februari 2024 · 11 min

Afspelen

Basic tips on how to assess inbound ML content and cultivate your news feed.
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/making-a-ml-feed Lees meer

00:00 How I assess all these AI releases
01:22 1. Model access and demos are king of credibility
02:31 2. Focus your feed on depth or breadth
03:09 3. Examples of using the model normally show its usable, shockingly
04:10 4. Leaderboards as the single leading claim is often anti-signal
05:00 5. Basic deep learning conceptual checks will often save you
06:13 6. If it's not even remotely reproducible or verifiable, it's not science
07:10 7. Don't over-index on Twitter
08:32 8. Data sharing, licenses, communication clarity, and small things add up
08:58 9. Research papers, technical reports, blog posts, and Tweets all serve different purposes
09:49 10. Socialize your information and build relationships

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

Google ships it: Gemma open LLMs and Gemini backlash

22 februari 2024 · 17 min

Afspelen

Google rejoins the open model party and gets some backlash for a frequent problem for generative AI.
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/gemma-google-ships-it Lees meer

00:00 Google ships it: Gemma open LLMs and Gemini backlash
03:12 Getting to know Gemma
07:11 Alignment details
08:55 Aside: What is REINFORCE? Some history of RL
11:08 Implementation details and RLHF
12:18 Terms of use: RAIL Licenses history repeated
14:05 Is Google back on top? Gemini's woes

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/gemma/img_008.webp
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/gemma/img_014.png
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/gemma/img_035.png
Figure 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/gemma/img_051.png
Figure 5: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/gemma/img_055.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

10 Sora and Gemini 1.5 follow-ups: code-base in context, deepfakes, pixel-peeping, inference costs, and more

20 februari 2024 · 15 min

Afspelen

10 Sora and Gemini 1.5 follow-ups: code-base in context, deepfakes, pixel-peeping, inference costs, and more
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/sora-gemini-follow-up Lees meer

00:00 10 Sora and Gemini 1.5 follow-ups: code-base in context, deepfakes, pixel-peeping, inference costs, and more
00:46 1. Deepfake detection of Sora
01:59 2. Playing with long-context, problem settings, and prompting
03:39 3. Gemini paper snooping: contamination and citation games
05:42 4. Training data and token estimates of YouTube
07:42 5. Unlocking model-based RL and downstream research
08:52 6. Midjourney style matching, V-JEPA, replicating Sora in the open
10:09 7. Architectures and academic links
10:57 8. Pixel peeping from the arts
11:58 9. Inference costs
13:24 10. Pressure on Llama and Mistral
14:03 11. Sound effects, physics, and the complete picture

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_003.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_007.mp4
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_009.mp4
Figure 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_011.mp4
Figure 5: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_037.mp4
Figure 6: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_044.png
Figure 7: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_047.png
Figure 8: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-2/img_049.mp4

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.interconne…

Releases! OpenAI’s Sora for video, Gemini 1.5's infinite context, and a secret Mistral model

16 februari 2024 · 9 min

Afspelen

Emergency blog! Three things you need to know from the ML world that arrived yesterday.
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/sora-gemini-and-mistral-next Lees meer

0:00 OpenAI's Sora for video, Gemini 1.5, and a secret Mistral model
0:53 Sora: OpenAI's text-to-video model
4:59 Gemini 1.5: Google's effectively infinite context length
8:01 Mistral-next: Another funny release method

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-gemini-mistral/img_015.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-gemini-mistral/img_023.png
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-gemini-mistral/img_026.png
Figure 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/sora-gemini-mistral/img_036.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

Why reward models are still key to understanding alignment

14 februari 2024 · 8 min

Afspelen

In an era dominated by direct preference optimization and LLMasajudge, why do we still need a model to output only a scalar reward?
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: In an era dominated by direct preference optimization and LLM-as-a-judge, why do we still need a model to output only a scalar reward? Lees meer

Podcast figures:
Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/reward-models/img_004.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/reward-models/img_009.png

0:00 Why reward models are still key to understanding alignment

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

Alignment-as-a-Service: Scale AI vs. the new guys

7 februari 2024 · 10 min

Afspelen

Scale's making over $750 million per year selling data for RLHF, who's coming to take it?
This is AI generated audio with Python and 11Labs. Music generated by Meta's MusicGen.
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/alignment-as-a-service Lees meer

00:00 Alignment-as-a-Service upstarts taking on Scale AI
04:25 The competition with humans-in-the-loop
06:05 Scaling Alignment-as-a-Service via AI feedback

Podcast figures:
Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/aaas/img_008.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

Open Language Models (OLMos) and the LLM landscape

1 februari 2024 · 9 min

Afspelen

A small model at the beginning of big changes.
This is AI generated audio with Python and 11Labs
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/olmo Lees meer

0:00 Open Language Models (OLMos) and the LLM landscape
6:24 Thought experiments
7:51 The LLM landscape heading into 2024

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/olmo/img_010.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

Model merging lessons in The Waifu Research Department

30 januari 2024 · 19 min

Afspelen

Note: some of the audio in the second half is a little wonky, but the general voice was upgraded so hopefully it's a little less "poppy" until then!
I'm trying to fix little pronunciation problems on a weekly basis. Thanks to my early fans! It'll keep improving. E.g. some of the months were wonky. Lees meer

When what seems like pure LLM black magic is actually supported by the literature.
This is AI generated audio with Python and 11Labs
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/model-merging

00:00 Model merging lessons in The Waifu Research Department
02:21 How and why does model merging work?
07:13 Aside: merging vs. ensembles vs. mixture of experts
08:21 Why are people doing this?
11:22 Tools & Links
11:51 Brief (visual) literature review
12:07 Full model merging and recent methods
15:55 Weight averaging during pretraining
17:18 LoRA merging
17:53 More background

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_005.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_016.png
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_042.png
Figure 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_051.png
Figure 5: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_055.png
Figure 6: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_058.png
Figure 7: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_060.png
Figure 8: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_062.png
Figure 9: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/model-merging/img_065.png
Figure 10: https://huggi…

Local LLMs, some facts some fiction

24 januari 2024 · 10 min

Afspelen

Local LLMs: the latency solution, Meta's open AGI, personalization myth, and moats X factor
The deployment path that'll break through in 2024. Plus, checking in on strategies across Big Tech and AI leaders.
This is AI generated audio with Python and 11Labs
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/local-llms Lees meer

0:00 Local LLMs: the latency solution, Meta's open AGI, personalization myth, and moats X factor
4:15 The personalization myth
7:13 Meta's local AGI and moats X factors

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/local-llms/img_026.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

Multimodal blogging: My AI tools to expand your audience

17 januari 2024 · 8 min

Afspelen

A fun demo on how generative AI can transform content creation, and tools for my fellow writers on Substack!
This is AI generated audio with Python and 11Labs
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/multimodal-blogging-tools Lees meer

0:00 Multimodal blogging tools
2:57 Stratechery, passport, and wonderful customer experiences
5:51 Wrap-up, features, and next steps

Figure 1: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/multimodal-blogging/img_006.png
Figure 2: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/multimodal-blogging/img_008.png
Figure 3: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/multimodal-blogging/img_012.png
Figure 4: https://huggingface.co/datasets/natolambert/interconnects-figures/resolve/main/multimodal-blogging/img_020.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

Multimodal LM roundup: Unified IO 2, inputs and outputs, Gemini, LLaVA-RLHF, and RLHF questions

10 januari 2024 · 16 min

Afspelen

A sampling of recent happenings in the multimodal space. Be sure to expect more this year.
This is AI generated audio with Python and 11Labs
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/multimodal-rlhf Lees meer

00:00 Multimodal LM roundup: Unified IO 2, inputs and outputs, Gemini, LLaVA-RLHF, and RLHF questions
02:46 Unified IO 2: Scaling multi-input, multi-output model pretraining
07:47 Collecting preference data for images
09:31 LLaVA-RLHF: The first experiments in multimodal RLHF fine-tuning
13:20 Multimodal RLHF questions, ideas, and resources

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

Where 2024’s “open GPT4” can’t match OpenAI’s

5 januari 2024 · 14 min

Afspelen

And why the comparisons don't really matter. Repeated patterns in the race for reproducing ChatGPT, another year of evaluation crises, and people who will take awesome news too far.
This is AI generated audio with Python and 11Labs
Source code: https://github.com/natolambert/interconnects-tools
Original post: https://www.interconnects.ai/p/open-gpt4-limitations Lees meer

00:00 Where 2024's "open GPT4" can't match OpenAI's
03:19 Models vs. products
04:51 RLHF progress: Revisiting Llama 2's release and potential in 2024
08:30 Smaller scale open RLHF
10:33 Opportunities
12:24 Commentary

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

Interviewing Tri Dao and Michael Poli of Together AI on the future of LLM architectures

21 december 2023 · 36 min

Afspelen

This interview is on YouTube and podcast players. Lees meer

Giving a voice to researchers is the best way to cut through the noise and understand what is happening with core developments of LLM technologies. I’m excited to get to talk with Michael Poli (Stanford PhD student + research at Together AI) and Tri Dao (incoming professor at Princeton + Chief Scientist at Together AI). This builds on the mega-post from yesterday on the same topics, though the interview is obviously less math heavy:

Interconnects is a reader-supported publication. Consider becoming a subscriber.

Introductions

[00:00:00] Nathan Lambert: Okay. Hey, everyone. Welcome to the first interview that we're going to post on interconnects. I'm really trying to bring more scientific voices into the AI discourse as media is covering a lot these days. I'm happy to be here with Michael Poli and Tri Dao, experts in some of these non attention architectures that have been really blowing up in the last few weeks of December.

So, Michael, do you want to introduce yourself first?

[00:00:25] Michael Poli: Sure. Thank you. Thank you, Nathan. For inviting me, I, do research at Together AI. And I was also a PhD student at Stanford, working with Stefano Ermon and, and, Chris Re, that's, that's how I met Tri as well. if I had to choose maybe, I moved to a few different areas of research.

if I had to choose one, I like to, do research at the intersection of signal processing, dynamical systems, and deep learning, and most recently, luckily, there's been more interest in, in kind of efficient architectures that use some of these principles. to improve scaling, along different axes and to, to get sort of new, new trade offs at inference time.

[00:01:13] Nathan Lambert: Great. And Tri?

[00:01:16…