bsdpython

Is it possible we hit a wikipedia moment (it being more accurate than Encyclopedias) where the quality of typical ai generated code is better than popular open source libraries?

Thanks. I know the last one is controversial, but the way I am starting to think about it is that we are just moving to a new layer of abstraction. I no longer understand very well how hardware works, nor do I know in detail how a browser renders a page, nor the full fine details of how many of the libraries I use work. My own AI generated code, in pockets, is starting to work in the same way. And I'm starting to become OK with that risk.

I did not get into it (very short write up), but in general there is a never ending need to build new software and systems. Those efficiencies will simply allow more to be done and more software to be built. That's been the history of software so far.

Google Cloud offers many different options to build Retrieval Augmented Generation (RAG) powered applications. This includes discrete components that comprise RAG solutions (embeddings, vector search, LLMs) but also includes options that combine multiple steps or even the entire RAG application in a single service. The best option for you will depend on factors such as your use case, engineering expertise, existing tech stack and future needs.

Let's start with a set of use cases and design a solution architecture using the most appropriate options. After that let's go through a detailed breakdown of the full list of services with the pros, cons and recommendations for when to use each.

Building this agent provides a nice lesson in why AI agents are so exciting but also so difficult to scale.

Besides the write up, you can try out the live demo (https://tinyurl.com/trendr-bot) and view the source code (https://github.com/brettdidonato/trendr-bot/tree/main).

I spent most of the past year presenting AI topics to executives, engineers and data scientists from companies that are Google Cloud customers. Last week was a new type of challenging audience: 2nd graders.

I spoke to three different 2nd grade classes at my 7 year old daughter’s elementary school. For the event I created a presentation and an AI powered website for the kids to create their own stories in a safe manner. The kids seemed to have a great time, with one kid even shouting out “I love AI!”

In the end, I directed students interested in AI to first focus on education and computer fundamentals: reading, writing, math, critical thinking and using computers to create things (but without using AI). This, along with getting the kids excited about technology in general, is the most important takeaway.

I have included a link to the presentation and code for the kids storytelling website within the article.

A short write up highlighting a few over hyped AI products. I’m starting to see a bit more thoughtful push back on the unrestrained hype, which is a good sign that this AI cycle is maturing.

How do you know which LLM is the best option to use for your particular use case? I published an open source repo to evaluate models based on your own set of prompts across Anthropic, Google and OpenAI. Besides model evaluation, it can also be useful for prompt engineering, API response time benchmarking and production application monitoring.

Sometimes I feel like I'm seeing something completely different from what is described in the popular narrative. This is a good example. I wrote a post detailing:

* What does Sora actually do? * What does it not do? * What will it likely be useful for? * And finally, what will be needed to actually replace the majority of video generation use cases?

https://shorturl.at/auIK0

Can We Prevent LLMs From Hallucinating? And if not, what implications does this have for the future of AI? Let's talk about it.

I pay for YouTube premium to turn off the ads. It's worth every penny.

Amazon also refuses to give feedback and then spam mails you to provide them feedback on the interview process. That tells you a lot about how they are as a company.

I spent 3 years as cto cofounder at a startup. After leaving I interviewed at lots of startups but I did not get much interest despite many having what I would consider a tech leadership vacuum. I ended up taking an offer from a FAANG and I am better off for now. This is in NYC and I have a degree and more varied corp experience so results may vary.

If you have a bunch of companies on your resume that no one has heard of then it doesn't help. If you are getting interviews though then there are other factors at play that I think you are dismissing. Are you fluent in in-demand skills? Do you do side projects? Did you do interview prep, especially coding exercises? What would you say are your biggest strengths?

A long overdue recession will solve many of these problems

It's a good list and these points are indeed very bad signals. When you break it down most points fall under the category of basic professional courtesy and it's shocking how rare that is. Most people that I interview and give prompt and honest feedback really appreciate it. You do have to deal with the occasional rude response however - it's the cost of being a professional.

"They tell you they like you, you’d be a great fit, your code looks great, but then discontinue because they just noticed you don’t have a piece of paper."

It could be that they somehow didn't notice and then singularly rejected you for that reason. It's also possible that they were willing to overlook it but once they had the chance to evaluate you as a whole they decided that the total package wasn't enough and the degree was the reason given. To me the degree is more about proving that the candidate has true interest in the subject and has been able to commit to it over an extended period of time more than anything else. I am asking them for a big commitment over an extended period of time to join my company after all.

"Internal reviews identified that the heart of the problem was the fact that the different design groups working on the project had used different Computer Aided Design (CAD) software to create the engineering drawings. The development of the aircraft was a collaboration between 16 sites spread across 4 different countries. German and Spanish designers had used one version of the software (CATIA version 4), while British and French teams had upgraded to version 5."

Really interesting! While I've hardly worked on anything as complex as an A380, I take it as a high priority for all software packages, from dev to test to prod, all be on the exact same version. Python 3.4 on one box and 3.6 on another? Let's not.

Being able to correctly identify logical fallacies is an important skill in and of itself. Persuading someone that they are wrong is a totally separate exercise.

Sequoia has 40 years worth of outliers. For almost any other fund I would agree but it's fairly typical for them.

There's simply not enough time to do all those things yourself. Find someone "good enough" in one of those areas and have them take over main responsibility.

I have a long list of features that I want to implement at some point. When I get feedback from users for those same features I move them up somewhat in my priority list. If it's something that's not on my list I add it to the bottom of the list and re-evaluate it's importance a few days later. I think it is a good way to avoid immediacy bias and keep a handle on what's a true priority.

People over process every time. It's really hard to build a team of good people though so instead companies try to tweak their process. Process can matter e.g. a good team can be wrecked by a bad process but a bad team is bad no matter the process.

I read the article why do you feel the need to summarize a part of it? Geez.

I don't like it but I do understand it from the employers perspective. Most hiring is done to avoid risk. Went to a top school, working at a popular company and already using the tech we use? Come in for an interview. Missing one of the three? We'll phone screen. Missing two out of the three? Not worth my time.

It probably has little to do with anti-China propaganda. Financial news has always reported junk like this with clickbait headlines and a few cherry picked numbers to make a story from nothing.

There are going to be a lot of mobile "full stack developers" that are going to have to scramble now to find another way to avoid learning how to build a simple API server and database.

I only used Parse for one small project a few years ago. At the time their scheduled tasks feature was new and I found it hilarious that there was a bug that mixed up hours and days in the scheduler. That is when the scheduled tasks actually ran at all.

I've been on both sides and I think it goes both ways. I won't take coding tests ever again because of the wasted time involved. And I've had cases where even existing consulting clients thought it was OK to try and cut my rate in half.

A few years ago I got publicity (TechCrunch among others) for a side project I developed. For at least 6 months I got non-stop solicitation emails for outsourcing and 30 second promotional videos. I can only imagine if I got that same publicity tied to a business with known funding.

A lot of these anecdotes are the frustration with dealing with people that waste your time. I guess it's not limited to just the startup space but one thing I've learned is that there is an unlimited number of people that are willing to waste your time and boy is it frustrating.

At it's core, a search engine is high tech while a social networking website is low tech. Before you want to argue, I know Facebook is nowadays damn high tech but that's not the point. Techies will always have more respect for a business built on actual high tech. See Apple. See Tesla. I happen to respect Zuck a lot for his amazing ability to grow the company and maintain control rather than his tech prowess, which is nothing to be ashamed of.