text
stringlengths
0
2.46k
[1628.66 --> 1630.90] I think it's the exact opposite of a zero sum game.
[1630.90 --> 1631.10] Okay.
[1631.10 --> 1635.48] I think there's like so much left to build that, uh, you know, the, the market is, is
[1635.48 --> 1637.08] huge and, and vastly growing.
[1637.88 --> 1642.00] Um, we do have, uh, features that copilot doesn't have.
[1642.00 --> 1647.28] So, you know, currently they don't have, uh, kind of like a chat based, uh, you know, textual
[1647.28 --> 1650.28] input to ask high level questions about the, the code.
[1650.98 --> 1654.16] Um, I think that's coming in copilot X to some extent, but
[1654.16 --> 1655.40] Yeah, I think they announced that, but it's not out yet.
[1655.46 --> 1655.76] I don't think.
[1655.76 --> 1656.38] It's not out yet.
[1656.38 --> 1659.26] And if you, if you look at the video, the, the kind of context fetching they're doing,
[1659.26 --> 1661.18] it's basically like, you know, you're currently open file.
[1661.26 --> 1662.00] Explain that.
[1662.00 --> 1664.30] And, and Cody is already doing much, much more than that.
[1664.30 --> 1668.14] It's, it's reading, uh, even if you ask it a question about the current file, it'll actually
[1668.14 --> 1671.98] go and read other files in your code base that it thinks are related and use that to inform
[1671.98 --> 1672.46] your answer.
[1672.46 --> 1676.48] So, so we think, you know, the power of source graph gives us a bit of a competitive edge
[1676.48 --> 1682.24] there with the kind of high level questions and onboarding and kind of like rubber ducking
[1682.24 --> 1683.14] a use case.
[1683.32 --> 1686.92] And then for completions, you know, I think copilot is, is great.
[1687.10 --> 1690.08] Um, but for, for completions, we're essentially doing the same thing.
[1690.08 --> 1694.92] So like the completions that Cody generates, it takes, uh, into account that same context
[1694.92 --> 1696.42] when it's, it's completing code.
[1696.42 --> 1703.60] So that means it's, it's better able to kind of mimic or, or emulate the patterns and best
[1703.60 --> 1706.42] practices in your specific code base.
[1706.42 --> 1711.88] And again, because we're kind of open source and model agnostic, we are just integrating
[1711.88 --> 1715.10] all the best language models as, as they come online.
[1715.10 --> 1718.94] So I think, you know, Anthropic, I don't know when this episode's going out, but Anthropic
[1718.94 --> 1720.36] today just, okay, pretty quick.
[1720.36 --> 1721.36] The 24th.
[1721.36 --> 1722.36] Yeah.
[1722.36 --> 1725.16] So Anthropic just announced today that they have a new version of Claude that has a,
[1725.16 --> 1728.24] a, an incredible, like a hundred thousand token context window.
[1728.24 --> 1733.60] It's just like, uh, uh, I think like orders of magnitude more than, than, uh,
[1733.60 --> 1735.28] what was previously available.
[1735.28 --> 1739.82] And, uh, that should be, I mean, by the time this episode goes online, that should, it should
[1739.82 --> 1740.88] be available on Cody.
[1740.88 --> 1741.88] Yeah.
[1741.88 --> 1745.82] Whereas, you know, Copilot, I think they're like, uh, maybe someone from GitHub can correct
[1745.82 --> 1748.94] me if I'm wrong, but I think they're still using the codex model, which was released in
[1748.94 --> 1750.10] like 2021 or something.
[1750.10 --> 1756.54] Um, and so it's a, it's a much smaller model, um, that only has around like 2000 tokens of,
[1756.54 --> 1760.38] of context window and much more basic context fetching.
[1760.38 --> 1763.46] It's already incredibly useful, but I think we're, we're kind of taking it, taking it to
[1763.46 --> 1764.46] the next level a little bit.
[1764.46 --> 1765.46] Right.
[1765.46 --> 1767.46] So open source and model agnostic.
[1767.46 --> 1769.62] Open source, model agnostic.
[1769.62 --> 1773.14] We're not locking you in to like a vertical proprietary platform.
[1773.14 --> 1774.14] Proxy friendly.
[1774.14 --> 1775.14] Proxy friendly.
[1775.14 --> 1778.38] Uh, also enterprise friendly, you know, source graph.
[1778.38 --> 1783.92] We, we, we, we've made ourselves easy to use in both cloud and on premises environments.
[1783.92 --> 1788.18] So we're just trying to do the best thing for our customers and for developers at large.
[1788.18 --> 1793.66] So because you're model agnostic, does that mean that you're not, you're not doing any
[1793.66 --> 1796.68] of the training of the base layer models?
[1796.68 --> 1798.42] So do you also sidestep legal concerns?
[1798.42 --> 1803.78] Cause I know like with, with codex and copilot, there's been, there's at least one high profile
[1803.78 --> 1804.94] lawsuit that's pending.
[1804.94 --> 1808.08] Like there's, there's legal things happening.
[1808.36 --> 1809.54] There's going to be things litigated.
[1809.82 --> 1813.32] I'm wondering if you're in the target for that now with Cody or if you're just not because
[1813.32 --> 1814.58] there's other people's models.
[1814.58 --> 1819.44] No, we're, we're very mindful of that and, um, we actually integrate models in a couple
[1819.44 --> 1820.00] of different ways.
[1820.00 --> 1822.56] So we do it for kind of like the chat based autocomplete.
[1822.56 --> 1826.46] There's a separate model we use for code completions and, and there's another model that we use
[1826.46 --> 1829.94] for like embeddings based, uh, code search and information retrieval.
[1829.94 --> 1832.40] Um, and it's kind of like a mix and match.
[1832.40 --> 1835.64] Like sometimes we'll use like a proprietary off the shelf model.
[1835.64 --> 1837.90] Other times we'll, we'll use the model that we fine tuned.
[1837.90 --> 1842.82] Um, but for the ones that, uh, the models that we do rely on external service providers
[1842.82 --> 1849.62] for, um, we're very mindful of the kind of like evolving legal and IP landscape.
[1849.62 --> 1852.94] And so one of the things that we're, we're currently building is, is basically like copyright
[1852.94 --> 1855.76] code, uh, or, or copied code detection.
[1855.76 --> 1859.78] And if you think about it, like source graph as a code search engine is, is kind of like
[1859.78 --> 1862.50] in, in a great position to, to build this feature.
[1862.50 --> 1867.30] It's like, if you emit a line of code or you write a line of code, uh, that is, you
[1867.30 --> 1874.04] know, verbatim copied from, uh, uh, somewhere else, uh, in open source or, or even in your
[1874.04 --> 1877.14] own proprietary code base, you know, you might be worried about just like code duplication.
[1877.14 --> 1880.74] We can, we can flag that for you because we've built, we've been building code search for
[1880.74 --> 1881.74] the past 10 years.
[1881.74 --> 1882.74] Yeah.
[1882.74 --> 1883.74] Cool stuff, man.
[1883.74 --> 1887.74] So moving fast, what comes next?
[1887.74 --> 1892.00] When are you going to drop Cody two?
[1892.00 --> 1893.76] It's probably like a week from now, right?
[1893.76 --> 1894.76] Yeah.
[1894.76 --> 1895.76] That's a great question.
[1895.76 --> 1899.72] I mean, we are just kind of like firing all on all cylinders here.
[1899.72 --> 1903.74] We have a lot of interesting directions to explore, like one, one direction or one dimension
[1903.74 --> 1906.68] that we're expanding in is just integrating more pieces of context.
[1906.68 --> 1911.68] So one of the reasons why we wanted to open source Cody, um, is because we just want to
[1911.68 --> 1916.24] be able to integrate like context from wherever it is and not be limited by, you know, a single,
[1916.24 --> 1917.86] you know, code host or a single platform.
[1917.86 --> 1922.66] Like there's so much institutional knowledge, uh, that's in many different systems might
[1922.66 --> 1923.66] be in Slack.
[1923.66 --> 1925.58] It might be in, uh, you know, GitHub issues.