

In this episode, we explore TabbyML's recent funding success and its ambition to create an open-source alternative to GitHub Copilot. The conversation delves into the implications of this development on the coding community and the future of AI-assisted programming.
Written by AI. May contain mistakes.

Get the gist, explore an idea or ask about something said.
On AI Today, what are the key points of this episode?
Automatic transcript. May contain errors.
0:00The AI-assisted code generation sphere is experiencing a burst of competition and startup Tabby ML, which is the brainchild of former Google employees is making waves. Recently, the startup secured, as I mentioned, over$3 million,$3.2 million in seed funding to kind of further enhance its open source code generator. So differentiating from GitHub's co-pilot, Tabby ML offers a kind of unique proposition as a self-hosted coding assistant. So Men Zhang, one of the co-founders, emphasized the tool's robust customization capabilities, saying, quote, We foresee a landscape where the customization and software development will be a predominant need for all businesses.
0:40I think while proprietary software solutions may have matured, when juxtaposed with GitHub's OpenAI-fueled tool, open-source solutions like TaviML kind of stand out. So according to Zhang, he believes that they stand out a lot. This is particularly evident in larger enterprise settings. Lucy Gao, Zhang's co-founder, pointed out how engineers in these environments often utilize proprietary code. And this makes tools like Copilot less effective, whereas open-source solutions like TabbyML can actually kind of thrive. So Gao illustrates this with a bit of a straightforward example saying, quote, if an employee pens a code line, I can seamlessly pull it using TabbyML.
1:19However, as with many AI innovations, code generators can have their shortcomings. So sometimes delivering buggy outputs, Gao however sees this challenge as a as fairly manageable for self-hosted platforms like theirs, as users make alterations to TabbyML's auto suggestions or dismiss them the tool learns and actually improves so if it's saying like hey here's a great bit of code for what you're working on and you say no that's not relevant it's actually going to learn from that and get better which i think is very interesting so the primary objective of these ai driven code generators isn't to you know necessarily oust human programmers but to kind of complement their efforts for now i'm sure in the future these things will be able to just completely automate a lot of these tasks and oust them.
2:05It's kind of funny, a lot of people talk about the fact that AI is not going to replace everyone's jobs, it's just going to help, you know, people that like augment everything you're doing. Well, yeah, but also I feel like it's going to replace a lot of jobs. And if you don't, if you're not really aware of that, or if you're not willing to admit that, then it's going to just be more disruptive when it actually happens. So I think in any case, a recent survey by GitHub highlighted that Copilot's recommendations were accepted by users at a 30 % rate. So further, Zhang kind of spotlighted an intriguing, you know, statistic from a Google developer gathering nearly a quarter of the tech giant software engineers and encountered over five assistive instances daily via its AI enhanced internal code editor, which is called CIDR.
2:48So launched just a few years ago, Tab EML has already gained, I think, some solid attention they have over 11 000 github stars um and you know one of the partners and zoocap have been identified as the primary investors in this funding round actually i think that was young key partners and zoocap um and i think addressing this kind of looming competition with the behemoth uh co-pilot of course where the big showdown is going to happen yang speculates that OpenAI's advantage might start to slip as AI models evolve and the expanses and all the expenses kind of associated with computing power decline.
3:27So currently GitHub and OpenAI's edge kind of comes from their ability to roll out AI models hosting tens of billions of parameters via cloud networks. Although there are, you know, costs to deploying such large models, Copilot has ingeniously balance this out to some extent by, you know, batching requests. So, you know, regardless of that, I think the tactic has its limitations. Microsoft reportedly bled over$20 per GitHub copilot user monthly in the initial months of this year as disclosed by the Wall Street Journal. So definitely not profitable and costing them a ton of money. I think in contrast to that, Tabby's strategy is to reduce development hurdles by endorsing models that train on one to 3 billion parameters so while this may compromise quality in the interim zhang is fairly optimistic about this and says quote as computing power becomes more affordable and open source models elevate in caliber github and open ai's competitive advantage is bound to ebb so i think it's going to be interesting to see if he's right about this or not definitely a very interesting area to watch
In this episode, we explore TabbyML's groundbreaking $3.2 million raise to develop an open-source alternative to GitHub Copilot, discussing its potential impact on the coding community and the future of AI-driven programming assistance.
Get on the AI Box Waitlist: https://AIBox.ai/
AI Facebook Community: https://www.facebook.com/groups/739308654562189
Podcast Studio AZ: https://podcaststudio.com/mesa-studio/
Podcast Studio Network: https://PodcastStudio.com/
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.