text
stringlengths
0
2.46k
• Safety concerns and the potential for future dangers
• Efforts to improve AI interpretability and understanding of model behavior
• Research on developing language models for non-English languages
• Using a public Discord server as a platform for discussion, collaboration, and research participation
• Involvement of volunteers and researchers from various organizations in AI research
• Discussion of the difficulty in creating effective safeguards against AI systems that are not very successful
• Mention of the speaker's own work in AI and open-source research
• Thanks and closing remarks from the speaker and the host
• Promotion of a premium subscription to the show's content
• Announcement of the show's conclusion and the host's promise to return on Friday
[0.00 --> 9.72] Welcome back, friends.
[9.72 --> 14.22] This week on the change law, we're taking you to the hallway track of the Linux Foundation's
[14.22 --> 18.06] Open Source Summit North America 2023 in Vancouver, Canada.
[18.50 --> 24.02] This episode is part of our Maintainer Month celebration, along with GitHub and many others.
[24.38 --> 27.44] Check it out at maintainermonth.github.com.
[27.44 --> 32.90] Today's Anthology episode features Byung Liu, co-founder and CTO at Sourcegraph,
[32.90 --> 41.44] Danny Lee, developer advocate at Databricks, and Stella Biederman, executive director and head of research at Eleuther AI.
[41.92 --> 46.08] The common denominator of these conversations is open source AI.
[46.72 --> 51.76] Byung Liu and his team at Sourcegraph are focused on enabling more developers to understand code,
[51.76 --> 59.34] and their approach to a completely open source model agnostic coding assistant called Cody has significant interest from us.
[59.76 --> 63.54] Danny Lee and the team at Databricks recently released Dolly 2.0.
[63.92 --> 71.56] This is the first open source instruction following LLM that has been fine-tuned on a human-generated instruction data set
[71.56 --> 74.94] and is licensed for research and commercial use.
[75.52 --> 82.82] And Stella Biederman gave the keynote address on generative AI and works at the base layer doing open source research,
[83.26 --> 84.92] model training, and AI ethics.
[85.34 --> 91.00] She trained the Eleuther AI Pythia model family that Databricks used to create Dolly 2.0.
[91.46 --> 98.22] A massive thank you to our friends at GitHub for sponsoring us to attend this conference as part of Maintainer Month.
[98.22 --> 115.86] Okay, before the show kicks off, I'm here with one of our sponsors at DevCycle, CTO and co-founder Jonathan Norris.
[116.30 --> 121.50] So Jonathan, my main question, I guess, if I'm handing off my feature flags to you all,
[121.74 --> 125.02] is my uptime dependent on your uptime?
[125.12 --> 127.66] Like, if you're down, am I down?
[127.66 --> 131.46] We've designed into all the SDKs and all the APIs.
[131.66 --> 132.56] APIs fail, right?
[132.64 --> 134.42] That's a cardinal rule of the internet.
[135.22 --> 140.84] So all the SDKs have been designed with kind of defaults and caching mechanisms and all that stuff in place
[140.84 --> 146.94] so that, yeah, if our CDN is down or our APIs are down, it'll sort of fall back to those defaults
[146.94 --> 149.32] or those cache values in those SDKs.
[149.34 --> 151.62] So that handles for those blips pretty easily.
[151.70 --> 156.78] And then we rely on Cloudflare as our sort of main high-load edge provider.
[156.78 --> 161.90] So all of our edge APIs are through Cloudflare and they're also operating as our CDN for assets.
[162.24 --> 166.78] So obviously relying on a large provider like that that runs such a large percentage of the internet
[166.78 --> 171.66] means that, yeah, you're not relying on our ability to keep AWS instances running properly.
[172.00 --> 176.06] You're relying on sort of Cloudflare and ability to sort of make sure the internet still works
[176.06 --> 178.48] as they control such a large percentage of it.
[178.48 --> 183.94] So yeah, we've architected it in a way that it doesn't sort of rely on our APIs to be up all the time
[183.94 --> 187.82] and our databases to be up all the time to have that good reliability.
[188.46 --> 189.14] Well, that's good news.
[189.38 --> 191.52] Okay, so how do you accomplish that?
[191.88 --> 196.40] One of the core sort of architectural decisions we made with our platform when we designed it
[196.40 --> 203.00] was trying to move the decisioning logic of your feature flags as close to the end user and end device as possible.
[203.00 --> 209.18] So we did that with those local bucketing server SDKs that are using sort of a shared WebAssembly core.
[209.56 --> 216.14] And then we have edge-based APIs that are also powered by WebAssembly to serve sort of those client SDK usages.
[216.28 --> 218.30] So things like web and mobile apps.
[218.50 --> 224.40] So that's one of our core principles is to try to get that decisioning logic as close to the end device as possible.
[224.56 --> 227.78] And this is probably one of the only use cases where performance really matters
[227.78 --> 232.38] because you want your feature flags to load really, really quickly so you can render your website
[232.38 --> 234.66] or you can render your mobile app really quickly.
[234.88 --> 238.62] And so, yeah, we definitely understand that your feature flagging tool needs to be fast
[238.62 --> 240.70] and needs to be really, really performant.
[241.22 --> 244.28] So if you want a fast feature flagging tool that's performant
[244.28 --> 246.88] and is not going to impact your uptime,
[247.26 --> 248.78] check out our friends at DevCycle.
[249.44 --> 252.74] That's devcycle.com slash changelopod.
[252.74 --> 256.98] And for those curious, they have a free forever tier that you can try out
[256.98 --> 260.48] and prove to yourself and your team that this is going to work for you.
[260.80 --> 264.94] So check it out, devcycle.com slash changelopod.
[265.40 --> 266.30] And tell me I sent you.
[282.74 --> 285.64] So, Cody.
[286.30 --> 287.10] Yeah, Cody.
[287.20 --> 287.48] Cody.
[287.80 --> 288.36] This is a big deal.
[289.36 --> 290.16] We think it is.
[290.20 --> 290.70] Seems like it.
[290.70 --> 290.82] Yeah.
[291.70 --> 297.42] Wasn't it Sourcegraph 4.0 last year was relaunched as the intelligence platform?
[297.60 --> 297.80] Yep.
[297.88 --> 298.30] Is that right?
[298.52 --> 302.08] Because before, not just, but just code search, which was cool,
[302.20 --> 305.08] but hard to really map out the ecosystem.
[305.30 --> 309.24] And you want all the space in there, but there was a limit to code search.
[309.24 --> 312.14] And you had to expand the insights and the intelligence.
[312.38 --> 315.14] And now, obviously, Cody is just like one more layer on top of insights.
[315.36 --> 315.84] Yeah, totally.
[316.08 --> 321.12] So, as you know, Sourcegraph historically has been focused on the problem of code understanding.
[321.38 --> 321.52] Right.
[321.64 --> 326.76] So heavily inspired by tools like CodeSearch inside Google or TPGS inside Facebook.
[326.88 --> 327.08] Right.
[327.20 --> 330.16] These kind of systems that indexed your company-wide code base
[330.16 --> 334.42] as well as your open source dependencies and made that easy to search and navigate.
[335.92 --> 339.02] And that's what's been powering the business for the past 10 years.
[339.02 --> 343.22] This is actually, you know, the 10th year of building Sourcegraph.
[343.42 --> 344.58] I was just wondering about that.
[344.58 --> 344.74] Wow.
[344.74 --> 347.86] Because when we first met you, it had to be about a decade ago.
[348.16 --> 348.30] Yeah.
[348.42 --> 352.84] I think Sourcegraph just either didn't exist or just had existed.
[353.24 --> 354.42] Sourcegraph existed when we met.
[354.66 --> 355.48] This was like GopherCon.
[355.48 --> 356.58] I think it was like 2014.
[357.28 --> 358.04] The first or second GopherCon.