dbish

26 ms·

331.
by dbish 3y ago
Sagemaker is probably a bad example since most folks who run ML workloads on AWS don't use it (from what I've seen talking to many ML teams). It's partially a scattered focus on their customer type (are they for experts/
332.
by dbish 3y ago
Having recently completed fundraising for a pre-seed, it's wild to me that I had VCs spending multiple weeks sometimes on "due diligence" for an idea-stage product, yet there are many examples of just yolo huge investments. I
333.
by dbish 3y ago
Eh, the bureaucracy was creeping in many years ago, RTO has nothing to do with that imho. Amazon has never really cared what's good for employees, and that's both a strength and a weakness, since it allows them to decide what&#x27
334.
by dbish 3y ago
As an ex-amazonian myself I agree with the first part about data driven and writing culture and that much of the way this is presented is very non-amazonian, but I'm not getting how rto is a "cookie cutter" example of day 2 b
335.
by dbish 3y ago
This is also why I’ve been recommending people to focus on small company roles. The big techs have too many internal folks who may have their job cut, and they also have recruiting orgs (mostly still) that are willing to burn your time with
336.
by dbish 3y ago
Having worked across multiple of the FAANGs, I would not trust joining them right now. They have not shown they will stop layoffs and morale has to be dropping as they each (most of them) do multiple rounds instead of the normally suggested
337.
by dbish 3y ago
I worked on both sides. I mostly agree except there are cases of important projects including AWS (like some of the ML work), also the whole aws usage discount/pricing thing is pretty huge and clearly the value in being within the same
338.
by dbish 3y ago
It’s still all Amazon, the single publicly traded company. Legal shenanigans/optimizations don’t change that. The other commenter was referring to AWS the org over Amazon Retail or Devices (other orgs).
339.
by dbish 3y ago
It becomes a ranking problem in a sense. Lots of data that can be the answer and lots of context that “could” be relevant to put into the context window but then you have to pick the right context and answer with the most correct informatio
340.
by dbish 3y ago
MSR seemed like it had a similar underlying purpose (or at least nice side effect).
341.
by dbish 3y ago
Exactly. Very hard to pin down any real inventions but very good at getting paid for the mystique
342.
by dbish 3y ago
Many (large majority I’m sure) of LLM applications that are cropping up are just thin wrappers on openai’s models and not terribly useful, but if you’re a developer I recommend using copilot. I was skeptical of the coding value until I star
343.
by dbish 4y ago
Cloud credits are pretty normal right now for startups (see AWS who led the way here with their 100k of credits). It's not "something or nothing", since they're willing to give out credits making the bet that you are hoo
344.
by dbish 4y ago
Exactly. All the big ones have partnerships/allegiances or own their cloud already, look at OpenAI and Microsoft or Google Bard on top of GCP, and similarly Stability seems to have a strategic partnership with AWS. You won't run a
345.
by dbish 4y ago
Very much doubt #1, the whole reason this works is model size. Smaller models and some intent model on top of it are very unlikely to produce the results they've been able to (and goes against everything they've shared on how this
346.
by dbish 4y ago
Interestingly this is shortly after another huge “seed” round in this space from fixie.ai (17mm). A lot of money being thrown at making it easier to chain/give LLMs access to other applications and tools.
347.
by dbish 4y ago
Agreed. Maybe I’m too focused on ML, but my first thought was the vector db. Cool idea though. Personally I use flask/quart to write the backend in Python but still end up writing a lot of JavaScript so this could be fun to use for pro
348.
by dbish 4y ago
Augmend | Software Engineering + Machine Learning Engineering/Applied Scientist | onsite Redmond/Seattle, WA | Full-time jobs: https://augmend.com/careers , company: https://augmend.com Augmend is devel
349.
by dbish 4y ago
I like Yann and think he’s brilliant and also generally appreciate how he engages on Twitter. However, he spent a lot of time recently downplaying openai and I don’t see why that was needed other then to try to say FAIR is important, see on
350.
by dbish 4y ago
It reminds me of a blander Yann Lecun-like response where he responded on Twitter to a bunch of people saying Facebook ai could have done this and that, without acknowledging they were beat to the (shipping) punch and are now just reacting.
351.
by dbish 4y ago
I worked on Alexa (and Cortana before that). I’m aware of the current tech. The tech barriers for doing this at high accuracy cheaply are still very much there.
352.
by dbish 4y ago
Having worked in big tech for a while, my written-by-a-committee alert went off quickly for this. Founders are the only ones who seem to get away without doing this when running large companies. Personality and substance are very important
353.
by dbish 4y ago
That would also be crazy expensive and hard to do well. They struggle with current speech reco that’s relatively simple, and can’t do this more complex always listening thing at high accuracy and identifying relevant topics worth serving an
354.
by dbish 4y ago
The email thing is company by company (Google seemed particular bad). Every company is doing it differently but from what I’ve seen/heard I’d say there are a few factors and the weight and use of these factors is based on company and e
355.
by dbish 4y ago
Love scrabble and tried upwords but the constraints aren’t right and it’s not as fun
356.
by dbish 4y ago
Happy to share. My main reason was that I really wanted to get more into machine learning and didn’t feel like it was going to be as fast to go deep on my own. It’s much easier now, both in terms of educational materials available and the a
357.
by dbish 4y ago
You can’t include “all” the context due to the prompt token length limitations. You have to use techniques for surfacing the right context that would fit in the limited context window and sending that along. There are techniques that keep a
358.
by dbish 4y ago
You can do a master’s part time while working with the right schedule. I did that as a full time software engineer working 40-50 hrs/week still. I’d rather this then all our representatives being ex law school or econ degree types tryi
359.
by dbish 4y ago
Hopefully he gets through them, looks like he’s still in the undergrad portion of this plan “must complete undergraduate prerequisite courses before starting the graduate work which he expects to start by 2024.”
360.
by dbish 4y ago
Master’s in CS with a concentration in AI is a better description. Cool to see a representative actually learning about the tech instead of just having legal education which is sadly far too common (same focus as my MS in CS though that was

More ›