Exclusive Guest Video - Swyx on The Rise of the AI Engineer
Added 2023-12-12 13:04:57 +0000 UTCThe debut video for the Insiders Arena, where guest experts - hopefully including you! - will bring in insight on concepts in AI, trends, or simply a cool coding project. This video is from none other than Swyx, made for AI Insiders exclusively.
Swyx is a developer/founder/angel investor, GitHub Star, Stripe Community Expert and helped run the React subreddit for over 200,000 developers.
He'll explain why we need the new term ‘AI Engineer’, why it’s smart to be active on Discord channels and the challenges and opportunities facing AI Engineers going into 2024.
Comments
Hi Shaun, could i also get access to the course please? Thanks!
Simon Kral
2024-05-05 22:39:38 +0000 UTCAmazing talk, really well paced. This new AI engineer category makes sense mostly, but I also feel like it's a bit too easy to be its own job. And it will probably only exist for a few years before it gets automated away.
Phillip Lakaschus
2024-01-06 16:47:59 +0000 UTCExcellent points about the shift to Fire-Ready-Aim, and AI+Code >> AI+Language. Thank you Swyx! On SW3.0, looks like metaprogramming in languages (generate inline code for themselves during execution) will be key. Although Python, JavaScript have some metaprogramming support, some rising languages like Julia excel much more at it - code is represented as a data structure of the language itself (similar to Lisp). This makes transformations and generation of code quite intuitive.
Martin Dimkovski
2023-12-17 19:07:50 +0000 UTCi think openai or other might converge on good architectures around their LLM´s without the need for AI Enineers - just a short term phenomenon. actually all that is needed is the conversion from LLM to agent in a reasonable way. there is no need for a engineer since by achiveing that there doesnt need to be any further tayloring all that is needed is already done - so why should such a job be needed in the future? I mean talking about applying optimization like Alpha Fold or GNoMe is sth else but might also be automatable once agents are around
En1gm 4A
2023-12-15 10:47:22 +0000 UTCThis is an excellent video thanks. Interested to get your view on whether the AI engineer is, or overlaps with the role I am currently playing as a semi-technical person - the AI equivalent of a product manager or business analyst in an enterprise setting. In this capacity there's the need to understand the capabilities and limitations of the various models and available infrastructure to access them, how models and assistants can be daisy chained, where those models can add value in a business context, and also be able to speak the language of developers (are these "AI engineers" or am I?) and business people including senior leaders.
Andrew Walker
2023-12-15 04:08:10 +0000 UTCThis is a very good point, my point of view is biased as my close team hasn't worked much within Legal but I have heard indeed they're very careful with hosting client data outside the EU. I know our MDR solution has a Europe-only hosting clause with a lot of firms, to ensure cloud data never leaves the EU. I'm sure that hosted LLM solutions will start offering similar clauses, I bet the new Mistral API is fully hosted in Europe.
Donato Capitella
2023-12-13 19:28:19 +0000 UTCI missed that, joined now thanks!
Bojan Savic
2023-12-13 18:53:26 +0000 UTCLoved this, thanks for sharing! Big fan of AI Engineering concept - it's going to be a big skills area to develop/hire for over the coming months/years
Sean Betts
2023-12-13 16:50:08 +0000 UTCMaybe this is just want my team and I are working on at the moment but I find that there isn't as much conversation about the overlapping skill set required for AI Engineering that combines engineering skills (coding) + language skills (prompting). Depending on much one is pushing the foundational model to it's limits, GPT-4 in our case, finding talent that can easily figure out how to stitch all of this together while not overburdening the model is an incredibly difficult task. I've spoken to other companies who are experiencing the same thing - They see the potential but their engineers can't quite find the prompting sweet spot. Some of the best software engineers I've worked with are finding this transition to a world of thinking about prompting and what LLMs can and cannot handle very challenging. To me it feels like the sweet spot is maybe having technical engineers who like to iterate/experiment but who also have great essay writing skills - that combination seems the winning set of ingredients. Maybe someone who did an Arts degree but turned to software engineer later in life might be a great example.
Andrew Thompson
2023-12-13 14:07:02 +0000 UTC@Donato, just to respond to "I do not see that much resistance now in adopting AI." I tend to agree with you holistically but I currently work in LegalTech and law firms are very cautious about their client data so much so that they're hyper conscious about any of it used to train models. OpenAI has said this doesn't happen unless you opt-in but I'm being asked for all sorts of guarantees around this. Also OpenAI say they store data for up to 30 days to prevent abuse, this seems reasonable to me but some law firms still have a problem with any of their client data being stored in the Us for even short periods of time.
Andrew Thompson
2023-12-13 13:54:27 +0000 UTCSeen chat on it in content discussion and elsewhere on the Discord but if my mods have to do anything more let me know
Philip
2023-12-13 08:28:13 +0000 UTCDiscord already launched! https://support.patreon.com/hc/en-us/articles/212052266-Getting-Discord-access
Philip
2023-12-13 08:24:28 +0000 UTCCan't come remotely close to the level Swyx is at but maybe when the discord launches we can have a channel focused on practical applications in production AI apps where we can all contribute our knowledge and then someone can use that body of knowledge for a talk.
Bojan Savic
2023-12-13 03:00:52 +0000 UTCcan we have a forum thread in discord to discuss this talk? i particularly want to talk abt the AI 3.0 and the LLM shell, code core vs LLM core, code shell https://twitter.com/swyx/status/1668679998566461440
SIM KIM SIA
2023-12-13 02:06:48 +0000 UTCis the slides for this available online?
SIM KIM SIA
2023-12-13 02:02:37 +0000 UTCSure of course: https://www.udemy.com/course/autogpt-gpt4-code-writing-ai/?couponCode=CB7981F8D6BE96979AC9 This should give you and anyone with this code a full discount for the next month I believe. Let me know if otherwise.
Shaun McDonogh
2023-12-12 20:26:31 +0000 UTCHi Shaun, could you please point me to your Udemy course?
Gautam Sabba
2023-12-12 20:10:01 +0000 UTCThank you Swyx! The idea of an "Ai Engineer" and Software 3.0 as AI + Code really resonates with me. Speaking to our clients, I do not foresee many organizations will want to train their own LLMs or even host them for the major part. The possibilities are more in leveraging and integrating LLMs into their own software offering via APIs run by big labs. I see something here that is possibly different than the previous tech "shift", which was about Cloud. At the on-set of Cloud, many organizations were really adverse to the concept, I remember the head of security of a bank saying just 8 years ago: "I do not see us ever going into cloud and hosting customer's stuff outside of our own datacenters". I do not see that much resistance now in adopting AI. Possibly the difference here is that the average decision maker could not "touch" the Cloud, whereas here everybody has been using ChatGPT et all - so that barrier to acceptance is lower. Many of our clients are now happy with using GPT-4 as part of Bing Enterprise and Office Copilot - after all, they are already using Office365, their private and confidential data is already in the cloud - so that risk has already been taken into account. I realize I went on a tangent, ultimately I can see how organizations will be willing to adopt LLMs via the big labs APIs in a much faster, natural and smoother way than it happened for the adoption of cloud.
Donato Capitella
2023-12-12 15:26:15 +0000 UTCFeel very aligned with the view of AI Engineer being a category unto itself. When building an AutoGPT that writes and executes its own code, I took inspiration from the Marvin GitGub project where we use the concept of AI functions to use hallucination to our advantage (btw happy to send you all free access to the Udemy rust course on building AutoGPTs if it’s helpful so just ask but don’t want to self promote here). Anyhow, what this really showed me is that automation is no longer just about inputs and outputs when writing code, but now there is a new factor, influencing an LLM which has emerged. Exhilarating.
Shaun McDonogh
2023-12-12 14:56:13 +0000 UTCMaybe you'll do one, one day?
Philip
2023-12-12 14:37:50 +0000 UTCStarting the new content category off with a bang! +1 sub for AIE
Bojan Savic
2023-12-12 14:26:23 +0000 UTC
