Відео

❤️

35:00 It sucks that researchers often let ego shape how they frame their conclusions. Synthetic data can be effective up to a certain threshold or boundary. While it isn’t a standalone solution for high-quality data provenance and model improvements, I personally believe there’s still considerable leeway before synthetically generated data becomes ineffective or hits that limit,

Moore Larry Young Donna Walker Jennifer

When you said, 'Look what it can do,' for a moment, I thought the guy on the bottom right was an AI replica of the one on the bottom left.

Jerde Wall

by far the best ai reality check interview

Seems like a solid founder but he seems low key desperate to communicate grammarly’s value. Too much of feature description, tell us why grammarly will matter.

this is great! or basically anything related to the future of creativity!

Thans for sharing guys

using AI to build AI...😁

Please somebody tell Jacob to speak in the mic :)

lmao ai

Introduction and Guest Introduction - 00:00:00 The Journey of Thomas D - 00:00:31 GitHub Co-Pilot and Its Impact - 00:01:03 Lowering the Barrier to Software Development - 00:03:36 The Role of Software Developers in the Future - 00:07:07 Co-Pilot Agents and Future Strategies - 00:10:02 Human vs. AI in Software Development - 00:11:59 Learning Journey for New Developers - 00:13:32 Impact of AI on Open Source Ecosystem - 00:17:21 Challenges and Surprises in Co-Pilot Development - 00:25:41 Future Milestones and AI Development Pace - 00:30:37 Impact on Different Parts of Development Stack - 00:33:26 Adoption of Multiple Models and Products - 00:38:25 Advice for AI Startups - 00:41:59 Importance of Focus in Growing Companies - 00:44:25 Extensions Program and Developer Tools - 00:46:13 Quickfire Section: Overhyped vs. Underhyped in AI - 00:51:09 Quickfire Section: Fears about AI - 00:51:48 Quickfire Section: Impressive Incumbent Companies - 00:52:23 Quickfire Section: Co-Pilot Surprises - 00:52:54 Quickfire Section: Exciting AI Startups - 00:53:55 Closing Thoughts and Wrap-Up - 00:54:27

but he doesn't know shit about it, it's not him and not the programmers who will decide the future of this technology, hasn't it sunk in yet?

This CEO is amazing. Super well spoken

I read: "github's CEO's baldest predictions..."

subbed

Keep up the good work, super excited to see your channel evolve. I loved the conversation. Ever since Runway was released I have been following the progress. I had never come across a video with the co-founder Cris so thank you for this video. All of the best!

WOW, Why this doesn't have views, you deserve it!

This is a great conversation,.

he sound so thoughtful and sincere

I was high all the time so they could rip me off 😊

Yeah! Cash reimburse ment?

why do they insistently refer to fine-tuning as pre-training, can someone explain?

All of this hype to create something that nobody gives a crap about.

Udio is better.

Contrary to the centralization of power; AI will decentralize the power into the hands of the masses. A decade ago, most people in the AI field openly said that AI will create new jobs as it replaces the older ones. However, most of those today admit AI can replace almost all human Jobs therefore "the power will concentrate in the hands of a few". However they forget that when all human labour becomes obsolete, all administrative "human" jobs will also become obsolete along with it, there will be no justification for those minority to stay in power, they will have the same power as everyone has, that is "one vote to one person". In short; this will be the "decentralization of power into the hands of the masses" that I am hinting about.

Just a clever path, that of Mistral, and a great episode. Thx

I've learned from this video that I'm not the only one who stares blankly and nods his head approvingly when someone says something at a level of complexity I can't grasp. I see others find that tactic useful too!

Wow! Beyang is very smart and charismatic. These are some insights 1. Being able to do a pull request is a good proxy indicator 2. They give you the option to choose which LLM you want 3. Their two main features are in line completion and chat 4. Context Fetching is one of the most important factors 5. They care more about product metrics over evals 6. Sometimes simpler is better when it comes to models like in the weather example 7. They have rate limiters to prevent abuse 8. Local models can sometimes be better due to privacy and cost 9. For developers, even shaving off milliseconds is important 10. They built the RAG internally due to abstraction boundaries/API necessary 11. One outcome of this whole GenAI revolution is that a company vertically integrates the platform and creates a proprietary developer platform.

I have to say I loved this interview ❤ just stepping into this ecosystem of AI, trying to understand and learn, there are so many at the helm of research and product development that openly shares their experiences and reflections in a very accessible way. Thanks so much!

Great podcast! Very technical point from both! I'm working for create the next big vector database with cool new features that i have in my mind.

An ordinary person (non-programmer) can try out Mistral in the Brave browser by clicking on the icon to the left of the wallet in the upper right-hand corner. A bit slow and limited, but it should be about the same functionality. 26 Class A investors in the company. It is free, so programmers can eliminate the bugs and compare its functionality to other LLMs.

its nice to see that there is competition

00:00:20 Start a conversation with your co-hosts about the future of AI policy. 00:01:00 Ask the guest about the origin of the company name "Mistral." 00:01:54 Discuss the decision-making process behind rolling out models via torrent and its reception. 00:02:24 Inquire about the story behind the company logo and its significance. 00:03:24 Explore the evolving landscape between closed source and open source AI offerings. 00:03:56 Explain the rationale behind having both open source and commercial offerings and the need to sustain open source development. 00:04:31 Discuss the strategy for deciding what features go into closed source versus open source offerings. 00:05:48 Explain the benefits of licensing technology to enterprises over licensing services. 00:08:00 Share insights on partnerships with companies like Microsoft, Snowflake, and Nvidia to facilitate adoption and distribution. 00:10:15 Discuss the preference of small companies for direct platform access and the facilitation of enterprise adoption through existing platforms. 00:11:51 Delve into the future frontiers of large language models, focusing on efficiency, controllability, and alternative architectures. 00:14:27 Analyze the impact of increasing compute power on model development and the challenges of transitioning to new architectures. 00:16:09 Evaluate the advancements in GPU technology, like Nvidia's latest chips, and their implications for training larger models. 00:16:52 Focus on evaluating AI products through a product safety perspective. 00:17:02 Consider the implications of the EU AI act and how it may evolve over time. 00:17:13 Reevaluate the approach to AI safety regulations to ensure product safety. 00:19:39 Rethink evaluation and continuous integration for AI products. 00:19:51 Provide developers with tools to verify that their applications work as expected. 00:19:59 Engage in discussions around transparency of training data sets while maintaining trade secrets. 00:20:03 Comply with regulations by offering evaluation tools and ways to improve models. 00:20:12 Encourage application makers to verify that their tasks are well solved. 00:23:21 Consider the importance of language and portability in deploying AI technology globally. 00:29:02 Address the challenge of acquiring new types of data for robust AI systems. 00:33:31 Prioritize solidifying documentation and guides to simplify the usage of APIs. 00:35:12 Focus on regulating applications rather than models to adapt to different regulatory regimes. 00:35:36 Consider the importance of transparency and training data in developing AI models. 00:36:22 Maintain flexibility in decision-making when building a developer platform. 00:37:00 Evaluate the need for partnerships in rapidly evolving environments. 00:38:38 Explore the potential of Foundation models for Material Sciences to drive advancements. 00:39:26 Share valuable content with others to spread awareness and knowledge. Generated by yousum.live

The mindscape is so refreshing.

why are these people celebrating their partnerships with these "hyperscalers". They will be consumed and consolidated. They are buying all the compute to rent back to you. wtf

1:15 "Ooh" wth was that 😂😂

People don't talk enough about MIstral and Arthur

Perfect!

thank you

WHat is the token linked to Mistral AI ?

Thanks for such a in depth convo 🙌

Perplexity is going to eat Google's lunch. So many ordinary people are sick of their manipulations and woke crap.

Perpelexity has great ui/ux and i have replaced google with perplexity ai

Google already has a gen ai output with its search

At 1.30, Those were 4 points and not 5

Amazing thought leadership guys. Thanks for bringing Clara to speak. I am a big fan of her for long time. Insightful AI outlook too!

I will be impressed if this unknown startup doesn’t go bankrupt in 2 years!!! 😢😢😢😢

I prefer perplexity than Gemini or Chat GPT.