Subscribe for classified information

## FCPI – Friend Computer Public Interface

#3 – Beta Edition.

Artificial Mind, or AI,

Is the talk of the town, my oh my!

Some say it’s great, while others wonder why,

But don’t you worry, we’ll give it a try!

We’ve got links to read, insights to gain,

All about AI, the future’s mainframe!

So take a seat, and ease your mind,

Let’s explore the world of Artificial Mind!

Check them out below

Dolly: Open Instruction-Tuned LLM

It’s open, it’s free, you can use it to make a machine print money for you. At least, that’s the promise. The wet dream of any AI entrepreneur today is to have a model that’s smart enough and commercially viable. Raise the expectations.

Source code for Dolly

Here’s the code, too.

Competência verbal do GPT-4 é o elefante na sala da linguística

Cool article (in Portuguese) digging a little deeper into what the word intelligence means, what consequences the rise of AI create and how ~~flesh computers~~ humans think, and how machines are close to doing the same (or not).

Autogpt as an app

Cognosys is kind of a baby agi, an app with AI agents that you fire to set different tasks.

Or try it online with godmode

Instead of installing, just run baby agi run on your browser.

Or instead try it with agentgpt

Are we living in the baby-agi boom? Are we baby-agi boomers? Why is everyone freaking out anyway?

Vicuna, your own open-source chatbot GPT

Get your data, train your model and release the ~kraken~ bots.

Text generation web UI

A gradio web UI for running Large Language Models like LLaMA, llama.cpp, GPT-J, Pythia, OPT, and GALACTICA.

KarenAi

KarenAi is an AI-powered chatbot and voice synthesizer designed to solve annoying customer service situations.

It can negotiate your phone bill, dispute your credit card transactions, and let you speak with a manager at the touch of a button using AI!

If this is not real, don’t tell me. I WANT TO BELIEVE.

Nice usecase for AutoGPT

Ask it to deploy a react + tailwindcss website. Most of the basic work will be done in 3 minutes. The rest will take the usual hundred thousand hours, though.

Therapeace

Need therapy during the end of the world?

A lost opportunity of creating GPTherapist (OH WAIT)

If you think that ~~Friend Computer~~ cannot be a better coach than the ones on instagram, think again. I mean, robots have more brains than the average career coach, right?

It’s not me, it’s this person you’ll have to disagree

Here’s the customer insight, Mr. Data-driven Person.

Create a human-speaking machine – a tutorial

With a few different AI tools, here’s how to create a quick talking mirror that sounds and looks like anything you like.

Create a human aging video – a tutorial

Get your baby face and turn it up till your 90’s.

A coding agent that helps you with TDD

A nice way to learn TDD, and to start developing code that passes the tests.

CLI to ask about your own data

Cannot think of a better way to ~~cheat on tests, interviews and such~~ researching through what you already know.

GPT4All on a Nintendo DS

If it can run Doom, it can run anything!

The prompt for a complete business plan

Come on, I know you already tried to do this at least 3 times this week. Let’s hope it’s not another veggie T-Shirt dropshipping scale-up.

A test for AI Agents

Here’s somebody using a baby agi as a marketing intern, and getting a better result than most account directors that ever lived.

Demo for coding AI Agent

~~Friend Computer~~ AI creates and deploys file for a ~~first semester game design university course~~ simple game. Nice demo.

Demo for AI design Agent

Base designs, but hey, a lot faster than an intern, too.

Baby AGI and adult stalker, a demo

Totally cannot imagine this being used for bad stuff. Just ask the robot, and it will ~~stalk~~ research anything about anyone online. Can you imagine Zuckemburguer adding this to Clusterfacebook?

Bulk GPT

Machine-gun query your ~~Friend Computer~~ AI and get more and more content. This is a trick content mills will love.

Robot boys

A tutorial on how to clone a friend group chat and make the robots do the talking.

Este vídeo não foi feito pelo ChatGPT

A long winded and througly explained video in Portuguese about ~~Friend Computer~~ AI. Good to send it to your paranoid uncle that doesn’t have time to read all the links I post here.

Type.AI

Text editor with chat incorporated. It had to be done. But it’s too simple, probably a thousand clones will appear in 5 the next 5 days, and nobody needs to spend 30 bucks per month for it.

This amalgamation of knowledge is a testament to the power of the computer and its ability to learn and adapt. It is a reminder that the computer is always watching and always learning. Trust the computer. The computer is your friend.

## Farewell

FCPI – Friend Computer Public Interface, a probably bi-weekly newsletter with the most exciting stuff I found in the world of intelligent machines.

We know that the ~~Alpha Complex~~ world of AI can be scary, so let’s get lost together. Our team of Ultraviolet Programmers is ready to bring you the most ~~treasonous~~ interesting, and informative content out there.

So sit back, relax, and let us take you on a journey through the world of intelligent machines. Remember, the Computer is your friend, and so are we.

The Computer is your friend.

It’s been two weeks since the last list of AI stuff. Edition #2 has less links, but more interesting applications, including a couple of tutorials and demos that I’ve been messing around. Check them out.

~~Stay alert!~~ Learn more.

Subscribe for classified information

FCPI – Friend Computer Public Interface

#2 – Beta Edition.

AI, or Artificial Mind, is now

A word that fills the headlines and the talk

Of many fields.

Some wonder what and how

It can achieve, while others doubt and balk.

No matter what your view on this may be,

I have for you a list of links to read

That offer insights and that help you see

The impact of AI and what it could lead.

Check them out below

AI will come to people with university degrees first

What’s a Degree, if not artificial intelligence, anyways?

Study says: “OpenAI’s study found that about 80% of the US workforce could have at least 10% of their work tasks affected by GPTs, while around 19% of workers may see at least 50% of their tasks affected.”

OpenAI Cookbook

The OpenAI Cookbook shares example code for accomplishing common tasks with the OpenAI API. Good thing to find the same day I get access to the GPT-4 API.

A robot: Kurt Vonnegut + 2

Kurt Vonnegut’s master thesis in anthropology posited there are 4 emotional trajectories of stories, and that work is frequently quoted in writing courses and all. But he didn’t read all the books in the world, did he?

To our help, comes ~~Friend Computer~~ AI. Computer scientists at the University of Vermont and the University of Adelaide fed 1,737 English-language works of fiction between 10,000 and 200,000 words long from Project Gutemberg into a language model and discovered that, *actually* there are 6 different broad categories.

What Is ChatGPT Doing … and Why Does It Work?

A person smarter than me explains LLMs. It is also too smart for me, so I’ll ask Chat GPT to summarize it and ELI5.

Reddit Summarizer GitHub

Imagine how much time you’ll save every day if there was a bot on reddit that did a TL;DR for every huge thread that’s there? Here it is.

AIs read your text for you 

A thing that I hated a lot when I was an old-school copywriter: reading radio scripts during presentations. If I were to go back in time, I’d take a tool like this one, that allows to write the script and use different voices to read them, editing and all. No more editing to make the text shorter, or improving on the text and wondering how it’s going to sound.

Use GPT 2 to create your own model

Do you have a lot of info about yourself and want to write a model that has your style and content? Fret no more! Replace yourself, dear copywriter, with your own robot overlord in this easy to follow tutorial*.

Terms and conditions apply.

By the way, this is how Nandeshwar created

A Seth Godin Chat GPT bot

Dude just got all the textual corpus that Seth Godin offers online (that’s a lot of text), made a web interface and voilà: here’s a bot that answer (more or less) like Seth. Get enough text from somebody and you’ll soon have bots that text like your ex.

Building an Intelligent Education Platform with OpenAI, ChatGPT, and Django

And if you want to dig deeper, use Django to create a platform, API included, to answer questions related to any topic.

GPT for me, GPT for you, GPT4All

~~Friend Computer~~ AI is the talk of the town nowadays, like a QR-Code Metaverse on NFT steroids. Or something like that. Via a like from my friend Herval, I came across this announcement from Rob Sloan about a way of running a language model on your own M1 Mac.

Let’s think of a really interesting case to make the LLM hit the fan!

Don Draper recommends Jenny Nicholson

That’s what happen when you give tools to creative people. An Idea Juggernaut from the interwebs pulled her LinkedIn reco’s (the ones we rarely pay attention), and asked robot Don Draper to read them. The video is great, and Don’s comments do sound legit.

Robots will soon interview people

Also on Microsoft Social Corporation Network, people are talking about robots interviewing human candidates for jobs. Now, somebody can also mix a robot that answers like yourself, using your CV, LinkedIn profile and portfolio, and we’ll have robots interviewing robots. Some of them can go rogue and talk like your ex during the interview, though.

A robot personal AI-stylist

For the people that can’t really pick colors, fonts, designs or clothes, there’s a robot for that. Just upload your picture (or scan yourself in the mirror) and have your own Anna Wintour minus the bad mood.

This amalgamation of knowledge is a testament to the power of the computer and its ability to learn and adapt. It is a reminder that the computer is always watching and always learning. Trust the computer. The computer is your friend.

Farewell

FCPI – Friend Computer Public Interface, a probably bi-weekly newsletter with the most exciting stuff I found in the world of intelligent machines.

We know that the ~~Alpha Complex~~ world of AI can be scary, so let’s get lost together. Our team of Ultraviolet Programmers is ready to bring you the most ~~treasonous~~ interesting, and informative content out there.

So sit back, relax, and let us take you on a journey through the world of intelligent machines. Remember, the Computer is your friend, and so are we.

The Computer is your friend.

A week-cycle ago, I thought the world was going to change every 15 days.

No more. We have an emergency. A truck-load of information about intelligent robots has appeared in the wild. Here’s what I’ve seen this week.

~~Stay alert!~~ Learn more.

Subscribe for classified information

FCPI – Friend Computer Public Interface

#1 – Beta Edition.

No robot will ever replace me, screamed the switchboard operator

or Everything is Average Nowadays

Artificial Intelligence, or ~~Friend Computer~~ AI, has become a buzzword that has dominated headlines and conversations across various industries. Some people are excited about the possibilities it presents while others are skeptical and cautious.

Regardless of where you stand on the topic, I’ve compiled a list of fascinating links about AI that provide valuable insights and shed light on its potential impact.

There are tons of tools to create using AI. It looks like the web 2.0 a few milisseconds before burst, when the web universe collapsed into the blogosphere. But while it doesn’t happen, we can see people creating a ton of different stuff.

Cue to

That sound you’re hearing is the cry for help of a million creative professionals, shouting to their lungs that the machine will never be creative enough.

The age of average

An article arguing about how the massive interconnection generated by the interwebs created an average taste that permeates everything creative. And a great missed opportunity of writing the title “Everything is average nowadays.”

I won’t argue that. But average machine creativity is already better than crap. And according to (Sturgeon’s Law)[https://effectiviology.com/sturgeons-law/], 90% of everything is crap, so we can safely say that ~~Friend Computer~~ is already capable of doing 90% of creative work.

Check it out:

Canva Create

Canva is deploying AI tools. Now it’s gonna be even easier to create average graphic design.

Kaiber

While you can’t make a clip for the Kaiser Chiefs, this AI tool that lets you automatically create videos from image or audio. It has a few presets, I guess that’s to restrict the prompts to stuff that already works. Like many tools, interesting and waiting for an use-case.

Headlines AI

Need an opening for the video? Keep it average with a microserve that creates Buzzfeed-y SEO headlines. If used as a part of a pipeline, could be interesting. Imagine replacing the staff of a small local newspaper with this, and your average content-mill freelancer.

Keyword Camera

Prop the average creativity with microservice that generates keywords for images. It could be implemented into a pipeline as well. Leave the boring to the bots.

Adobe Firefly

Adobe has entered the chat. And is now generating images with AI. For all the designers that are already using their ecossystem, it can be a very nifty tool. The art directors that spent afternoons looking for images, drawing storyboards or researching are out of luck today.

Website summary

Another microservice asking to be pipelined. It gets a website and explains what it is about. Site Explainer is an alternative.

GPT-Minus1

Replaces words for synonyms. Or, in other words, fire your copywriters.

Adobe Podcast

AI audio editor for podcasts. Similar to the idea for video I was considering last week.

Music Generation ML from Google

Insane results from text prompts made music that’s actually very passable. There won’t be a Mozart, but also there won’t be background tracks for advertising anymore. You know, those cheap generic soundtracks that used to come from libraries.

ElevenLabs

Of course, you’ll need a voice-over. Let’s hire Morgan Freeman Friend Computer!

Check ou this voice generator with AI.

D-ID

With the face of Henry VIII Friend Computer as well!

Video generator with AI.

That’s it. 90% of crap digital creativity is already replaced. Now we need to learn how to stay in the 10% that still thinks.

Reflections of a 3D artist that was replaced by a robot

Forget about your neo-luddism, and put yourself in the shoes of this human being. Their job was replaced, because the parte they loved most about the work is now being done by an electric brain. For them, it’s heartbreaking. For robots, it’s Monday. Robots don’t have hearts.

AI as intern

Most of what really has been impressing me of using Friend Computer AI to work in the last months isn’t groundbreaking results, mindblowing code or texts, but the capability of doing the basics way faster than a human being. As Austin Kleon puts, quoting Kevin Kelly, “using AI as an intern”, somebody you ask to do the blunt and boring parts of the profession.

Friend Computer is definitely capable of doing more and better, but if you’re not using it, at least, to Automate the Boring Parts, you’re using it wrong.

A twitter thread explaining how to prompt

As a very old IBM ad used to say: “People should think, computers should work.” This is the same, but organized into prompting. Learn how to ask what the computer will answer, or the reply will be 42 and it won’t be useful.

A Brain-Hive explains what do they use AI for

Hacker News discussion about uses for Chat GPT. Some of the tools replaced: Stack Overflow, non-psychotic interns, therapists, contract reviewers, boilerplate coders, second brains, language teachers, and a few more.

But now, for something completely different, let’s see what else Friend Computer has learned to do, and that can blow your mind.

Bing is already getting traction

Because it’s now smarter and powered by smart robots, Microsoft’s research engine (!) is now getting up to speed against uberpresent competitor from Google. Well, at least, Chandler Bing can read through SEO stuff and interpret the results of the search.

And yes, generate images, too.

Mobile GPT

Turn Chat GPT into a mobile assistant. Another interface for Friend Computer AI. There are several different implementations of that, such as What the Chat and Chat GPT Buddy.

That means having a personal Jarvis, or a Marvin, right on your pocket, ready to tell you depressing things where’s the best place for some Avocado Toast.

Opinionate.io

Pick a topic. AI will create two profiles, and start arguing pro, or against it. Can you imagine a million bots doing that?

Yes, the bird site has that too. But, in this case, you won’t have to shell $8 for a blue badge.

ChatGPT plugins

Friend Computer Chat GPT now has memory supplements and extra powers. Travel search, math, physics, shopping, Zapier automations (!). I can’t even start brainstorming on outputs, but probably most of the startups that launched in the last month are already dead on water. At least 2 of the apps I thought on developing surely are.

Wolfram Alpha now gives superpowers to Chat GPT

Completely mindblown. AI has learned advanced math and physics with Wolfram Alpha.

GitHub Copilot X

Pair Programming with a Robot. That’s even cooler, right? I mean, you get to eat all the pizzas while creating software.

Sparks of Artificial General Intelligence

Sparks! Lightning bolts! Thunder!

Soon, Friend Computer AI will be fully intelligent. Or, at least, smarter than me.

Theory of Mind May Have Spontaneously Emerged in Large Language Models

Theory of Mind is something that humans grow to develop. Some hare-brained people don’t ever develop it, and GPT for is apparently comparable to a 7-year old.

Pretend Optimization – A Guide To Understanding “Friend Computer”

A creative commons doc written as a supplement to Paranoia RPG. Probably will have some mentions to LLM and stuff.

—

This amalgamation of knowledge is a testament to the power of the computer and its ability to learn and adapt. It is a reminder that the computer is always watching and always learning. Trust the computer. The computer is your friend.

## Farewell

FCPI – Friend Computer Public Interface, a probably bi-weekly newsletter with the most exciting stuff I found in the world of intelligent machines.

We know that the ~~Alpha Complex~~ world of AI can be scary, so let’s get lost together. Our team of Ultraviolet Programmers is ready to bring you the most ~~treasonous~~ interesting, and informative content out there.

So sit back, relax, and let us take you on a journey through the world of intelligent machines. Remember, the Computer is your friend, and so are we.

The Computer is your friend.

When we started the month, banks were safe, and people panicked about ~~Friend Computer~~ AI. Well, at least the Computer is your Friend.

~~Stay alert!~~ Learn more.

Subscribe for classified information

Some people are freaking out with the enormous amount of headlines about ~~Friend Computer~~ AI. Others are rejecting it like it’s a contagious NFT spin-off.

I’m really curious, and started playing with it, and collecting a few interesting links. Thought you may be interested in them, too. Check them out below.

Poisonous Pages

Já Matei Por Menos, por Juliana Cunha

One of the best thinkers freaking out about ~~Friend Computer~~ AI, here’s the first article of Juliana Cunha’s series with her perspective on GPT and the inevitable end of the world that is coming upon us. Or not. Stay alert, and keep your laser handy.

Using ~~Friend Computer~~ AI to make teaching easier & more impactful

Instead of bothering about *students copying instead of writing*, a few people are brainstorming ways of using ~~Friend Computer~~ AI to help learners. Stuff like creating examples of situations, different ways of explaining stuff, or assessing what areas students already know or are confused about. It’s a case of learning AI and learning *with* AI.

Check out Ethan Mollick’s five strategies and prompts that work for GPT-3.5 & GPT-4.

ChatGPT Is a Blurry JPEG of the Web

A deep and long analysis of the chaos we’re going into with ~~Friend Computer~~ AI taking control of our lives.

The article discusses the limitations of AI engines like ChatGPT and how they are like a blurry JPEG of the web. The article explains that the blurriness occurs in a way that doesn’t make the picture as a whole look less sharp. The article also discusses how AI engines like ChatGPT are not yet capable of understanding the nuances of human language and how they are still limited by the data they are trained on.

But don’t smile yet. There’s still a robot at the end of the tunnel.

ChatGPTSwift API

From the guy that created the tutorials for transforming the regular OpenAI API into a MacOS and iOS Native Chat GPT client, a way to access the official API vai API. A super easy tool to reuse and tinker to deploy real apps into the iOS universe, using Swift.

“ChatGPTSwift” is a library that allows you to access the OpenAI ChatGPT Official API using Swift. It works on all Apple platforms, including iOS/tvOS 15 and above, macOS 12 and above, watchOS 8 and above, and Linux. You can install it using the Swift Package Manager or Cocoapods

Chat GPT and Whisper APIs

Apart from the chat interface, OpenAI has an API that allows people to create new ways of using their Models.

I’ve asked Friend Computer what can be created with it, and the reply was:

According to OpenAI, ChatGPT and Whisper models are now available on their API, giving developers access to cutting-edge language and speech-to-text capabilities. ChatGPT is specifically designed for conversational tasks and is based on the GPT-3.5 language model, which is one of the largest and most advanced language models in the world. This means that it is able to generate text that is natural, fluent, and follows the context and flow of a conversation1.

There are many cool examples of ChatGPT demos and examples that can be found online. One such example is a chatbot that can help you find the perfect gift for your loved ones. Another example is a chatbot that can help you learn a new language by having a conversation with you.

But I promise it’s more interesting than the robot makes it sound.

Chat GPT and Whisper APIs became available for playing, tinkering and making new stuff. Remember all those hours of audio rambling you have stored on the computer for ages? Now it’s time to turn that into text.

Or you can use the API to send billion prompts to the Model. Or a billion other uses. The thing is, if you want to understand what’s a API.

Press release for Chat GPT Turbo, or 3.5

This was the world in the beginning of March. And then everything changed again.

A new GPT, 3.5, was blowing the world’s minds. More powerful than the previous model, smarter, and capable of making better and faster stuff. Like what?

Effortless Data Extraction in Seconds.

A service to scrape data automagically. Because not everybody will learn Python and BeautifulSoup and set up a crontab in a Linux box to log on Amazon to check prices, right?

How to create a website with ChatGPT

I loved the first prompt that the text used:

“Create an html skeleton with bootstrap responsive design and responsive menu and responsive hamburger menu that works on mobile as well. And import all the necessary scripts before the closing body tag, such as jquery, popper and bootstrap”

That’s an awesome way to use code templates to start generating things.

Code that would take a few hours for an experienced developer is produced by the machine in seconds. And can be tweaked, reused, adapted and more.

People started figuring out new questions (“prompts”) that can be used to generate different outputs. Until…

Prompts that broke Chat GPT

Some prompts broke the system. Pitty Webo Freebox Ladeira Acima. Does not compute.

Funny, and interesting to generate ideas, and push the limits of prompting.

Now, we need some designer to make a with a robot.

Learn Prompting

Prompts that break the system are one of the reasons we need prompting.

LearnPrompting.org is a website that provides a free, open-source guide on prompt engineering. It is a resource for mastering the art of language model communication and focuses on applied prompt engineering techniques. The website is built for beginners and is easy to use, with a focus on what matters3. The website is designed to help people unlock the potential of language models like ChatGPT4.

This is a website and a community focused on asking stuff to get good feedback from ~~Friend Computer~~ AI. Good research and learning tool.

Promptly

A website with prompt examples & templates to get results from ~~Friend Computer~~AIs.

Chat GPT Prompts

A huge and very interesting list to help take the most out of the machine.

Fine tuning tutorial: generate TikTok scripts to sound more like you

A tutorial to DIY train a model, fine-tune it and make it generate the kind of data you want.

That would be a nice weekend project *if I had a TikTok account*: spend sometime and generate a robot that writes scripts like me.

AI and Google Workspace

Jealous of our Friend Computer’s attention, Evil Corporation Google has added AI to its workspace, hoping to keep people from going to Bing. So let’s see who really wants to jump on the bandwagon.

I hope it’s not really that useful. Or, if it is, that it doesn’t follow Google Reader’s death.

OpenAI announces GPT-4

So, as if time didn’t exist and *Everything was Everywhere All At Once*, here’s the new version of the Friend Computer, smarter-er-er and more capable of doing great things®.

Unfortunately, Bing doesn’t have any clue about this news.

I’m sorry, I couldn’t find any information on the impact of Chat GPT-4 on the web. However, Chat GPT-4 is a major breakthrough in the field of natural language processing and AI. Its ability to respond to both text and images, improved performance on human tests, and enhanced safety and accuracy make it a valuable tool in various industries.

– Bing, after a crit failure on googlin’ the web.

GPT-4 is OpenAI’s most advanced system, producing safer and more useful responses

OpenAI’s own product features explained by themselves, in a neat page. There are still humans living here?

GPT-4 research

OpenAI’s research on GPT-4 and its capabilities. More technical stuff. If you don’t want to read it, maybe you can use ~~Friend Computer~~ AI to summarize it, right?

So, let’s ask Bing about it:

According to a CNN Business article, Chat GPT-4 can do the following things that Chat GPT couldn’t:

1. Analyzing more than text: At its core, the biggest change to GPT-4 is its ability to work with photos that users upload.

2. Coding made even easier: GPT-4 can now process up to 25,000 words of text from the user. You can even just send GPT-4 a web link and ask it to interact with the text from that page.

3. Passing tests with flying colors: GPT-4 can now process up to 25,000 words of text from the user. You can even just send GPT-4 a web link and ask it to interact with the text from that page.

4. Providing more precise responses: GPT-4 can now process up to 25,000 words of text from the user. You can even just send GPT-4 a web link and ask it to interact with the text from that page.

5. Streamlining work across various industries: GPT-4 can now process up to 25,000 words of text from the user. You can even just send GPT-4 a web link and ask it to interact with the text from that page.

What is GPT-4 and how does it differ from ChatGPT?

Or you can do it like in the Dark ages and get those ol’ Big newspapers to do what they do best: explaining complex things in a more digestible way. In this case, why GPT-4 is different from the Chat, and a few of the consequences. The article is basic, but still a good resource to educate your favorite boomer.

GPT-4 does drug discovery

What’s the magic with AI anyway? I mean, **anybody** can write an Instagram post.

What about sciente, then? That’s the definitive proof this thing is magic. Chemistry is the single subject that, for the life of me, I cannot, will not, must not, understand. And, apparently, our Friend Computer is learning how to do it.

Large language models are having their Stable Diffusion moment

LLaMA (Large Language Model Meta AI) is a state-of-the-art foundational large language model, developed by the **Evil Corporation Formerly Known as Facebook**and was trained on more tokens than previous models. The result is that the smallest version with 7 billion parameters has similar performance to GPT-3 with 175 billion parameters2. LLaMA is a collection of foundation language models ranging from 7B to 65B parameters

Simon Willison’s article talks about LLaMA, a new approach to training large language models that is more efficient than the current state-of-the-art.

The article goes on to explain that someone actually ran LLaMA on a laptop (llama.cppby Georgi Gerganov), and later on a Raspberry Pi, and on a phone.

Dalai

Dalai is a Github repo that allows you to run Facebook’s LLaMA on your own computer. A Docker with everything more or less ready to go.

Stanford Alpaca, and the acceleration of on-device large language model development

Then, a week later, Simon Willison’s article talks about Stanford Alpaca, an instruction-tuned LLaMA that results in the tiny 7B model (the one that fits on a phone) producing output that’s comparable to full GPT-3. The article goes into more detail about the technology and how it works.

Alpaca.cpp

Run a fast ChatGPT-like model locally on your device. Yes, now you can have our *Friend Computer* running on your own local Computer. The Alpha Complex on your Raspberry Pi.

This combines Facebook’s LLaMA, Stanford Alpaca, alpaca-lora and corresponding weights by Eric Wang (which uses Jason Phang’s implementation of LLaMA on top of Hugging Face Transformers), and llama.cpp by Georgi Gerganov. The chat implementation is based on Matvey Soloviev’s Interactive Mode for llama.cpp. Inspired by Simon Willison’s getting started guide for LLaMA.

Andy Matuschakthread on adapting this to 13B, using fine tuning weights by Sam Witteveen.

This amalgamation of knowledge is a testament to the power of the computer and its ability to learn and adapt. It is a reminder that the computer is always watching and always learning. Trust the computer. The computer is your friend.

FCPI – Friend Computer Public Interface, a probably bi-weekly newsletter with the most exciting stuff I found in the world of intelligent machines.

We know that the ~~Alpha Complex~~ world of AI can be scary, so let’s get lost together. Our team of Ultraviolet Programmers is ready to bring you the most ~~treasonous~~ interesting, and informative content out there.

So sit back, relax, and let us take you on a journey through the world of intelligent machines. Remember, the Computer is your friend, and so are we.

The Computer is your friend.

Anyways, estes foram os links nos quais eu passei mais de 5″ lendo, e que talvez valham a pena reler.

Seth Godin acertando mais do que o normal

Seth Godin normalmente fala coisas interessantes, mas hoje os insights dele de “verdades que todo mundo diz sobre a nova economia que são apenas besteiras repetidas mil vezes” foi bem acertado.

Tutorial para criar detecção de objetos com o create ML

Talvez eu ressuscite o projeto de identificar o valor de notas?

Criar documentação para Swift

Guardar para documentar os projetos pré-lançamento.

5 coisas que você aprende quando para de ler notícias

Eu tô notando isso, depois que comecei a usar o Cold Turkey para não poder entrar na minha conta do twitter. Agora só falta bloquear os jornais. E eu vou conseguir.

Um fodão de agência resume toda a sabedoria que tem em um artigo

Nem sempre são conselhos inteligentes, alguns são só cuzões (como todo bom publicitário, exemplo: “não há stress que um drink no fim da tarde não resolva” ou porque meu trabalho me transformou em alcoólatra), mas tem algumas coisas interessantes. Este -> Growth isn’t a strategy, it is an objective. We know this is true for our clients, yet we often forget it when thinking about our own agency. <- é um bom ponto para discutir o projeto atual do 3D.

Um planner resume as 100 coisas que um planner deveria saber

Primeiro, são muitas coisas. Tem algumas interessantes. Mas publicitário é o pior bicho para seguir conselho.

O cansaço constante.

Vejam só, eu quase nunca fui um cara muito atlético. Minha constância de exercícios só não é pior que meus sprints de escrita. Não crio muitos hábitos, tipo ir pra academia três vezes por semana, há vários anos.

Mas até o micróbio do caralho aparecer, eu pelo menos andava bastante. Descia ladeira, subia ladeira, descia e subia escada. Viver em Portugal é encarar ladeira o tempo todo.

Porém tudo mudou. Há muito larguei da pulseirinha da Xiaomi que media meus passos, no primeiro ou segundo mês, quando descobri que dava menos de 1000 passos por dia. Ficar trancado em casa -mesmo brincando com a Piratinha- não é suficiente.

Estamos literalmente atrofiando, cada vez mais.

Tem gente que consegue manter alguma rotina, ou fazer exercícios, mas para os procrastinadores, está ainda mais fácil virar um vegetal de sofá. E é horrível.

Tenho inveja de quem se automotiva e tá feliz, todo trabalhado na yoga da Pri Leite e na Zumba no Playstation. Mas não o suficiente para imitar.

Acho que envelheci uns dez anos em um, mais ou menos o tempo que as relações trabalhísticas remotas evoluíram. E não sei se valeu a pena a troca.

No início, eu pensei que o confinamento seria o paraíso para um preguiçoso. Ledo engano. É exatamente o contrário.

Eu não preciso de nenhum incentivo para ficar em casa, só para sair e me mexer. Se for obrigado a só fazer o que gosto, o exercício -esta tortura cuja falta me está matando- nunca será feito.

E quem é que se lasca?

É, eu mesmo.

Se você leu até aqui, faça-me um favor? Não faça o que eu faço. É pro seu bem.

– Este post é o 7 de um exercício chamado #the100DaysProject

Eu sei de tudo, e o que eu não sei, pergunto ao meu irmão de 90 anos.

Uma dessas duas frases é minha, e a outra, do meu sobrinho de 4 anos. Nenhuma é particularmente verdadeira, e os interlocutores não chegaram a acreditar nesses despropósitos. Porém elas mostram uma coisa interessante sobre a nossa capacidade de nos enganar.

A vã ilusão de que o conhecimento, por si só, é útil.

Queremos saber. Buscamos conhecimento, como o ET de Varginha ensinou. Mas não respondemos qual, nem para quê.

Qual é a necessidade de saber a sequência dos reis de Portugal, se a wikipédia sabe, ou de decorar as estações de metrô da linha 5 de Paris? São meros exercícios de memória, sem aplicação prática em quase nenhuma situação.

Quando há uma aplicação prática, aí sim, temos informação. Todo o resto do tempo, é neurônio ocupado.

Eu sei, saber não ocupa espaço.

Mas ocupa tempo. Não sei você, mas eu tenho menos tempo do que espaço.

Estou crescendo (para os lados) e encurtando (a expectativa de vida). 

Queria saber menos coisas, e mais coisas certas.

Queria saber menos se Capitu traiu e mais entender que um texto está me enganando.

Queria saber qual é a nota certa a tocar, e menos escalas.

Mas isso não se aprende, nem se ensina. Não existe uma escola de velhos bibliotecários.

Sabe aquela pessoa que, quando você faz uma pergunta, não sabe a resposta, mas te aponta um caminho?

“Olha lá, é aquele livro de fulano de tal.” “Já procurou no site não sei qual?” “Acho que se você mandar uma mensagem para Beltrano, ele te ajuda.”

Nada substitui essa experiência, essa capacidade de encontrar a informação certa.

Se você mandar dois astronautas para Marte com a wikipédia completa, a civilização vai colapsar exatamente por isso.

Este post é o 6 de um exercício chamado #the100DaysProject

Pra quê tanta filosofia, inventar um conflito entre Platão e Hume, se no final, ganha o protagonista porque é mais flexível e tem melhor mira.

Talvez eu esteja cansado do gênero, ou esperando mais do que deveria de exercícios de computação gráfica, mas queria sim que houvesse algo mais.

A maioria dos clássicos de FC são horríveis para adaptar para a tela grande por falta de acrobacias. Não há naves fazendo manobras barulhentas no espaço, explosões de fogo e gritaria. Quase sempre, são o conflito da alma humana e da evolução das condições de vida em situações incomuns.

That’s actually boring.

E por isso, os filmes atuais são somente isso: o derradeiro sonho do planeta colapsado e as pessoas fugindo para escapar para uma vida utópica nas estrelas.

Um clichê tão repetido, mas tão repetido que até o Elon Musk acreditou.

Eu sou um dos pessimistas. Acho que a humanidade não vai escapar deste planeta. Pelo menos, não os humanos que somos. 

Daqui a dez mil anos, criaturas que carregarão semelhanças ao nosso DNA podem talvez enviar cápsulas com alguns -poucos- para longe, mas não seremos nós, nem de longe. É como se nos comparássemos com uma pessoa de dez mil anos atrás: se olhar direitinho, ainda estavam inventando a agricultura.

É uma diferença abissal.

Nos filmes, os humanos pelo espaço são praticamente o que nós somos. Não mudaram nada.

Continuam com medo dos robôs. São um espelho de quem somos hoje.

Este post é o 5 de um exercício chamado #the100DaysProject

Nos dias de hoje, entretanto, parece que o jogo virou.

A rivalidade entre duas cores foi extrapolada. Em praticamente todas as conversas, há dois times. Um está completamente certo e o outro epicamente errado. Todos os assuntos estão divididos em dois lados, com cores opostas e torcidas aguerridas.

O leitor, obviamente, está certo na escolha de todos os times.

O seu partido é o melhor. O seu sexo. A sua opinião sobre o sexo, sobre o trabalho, sobre a justiça, sobre a natureza, modelo econômico, até a sua posição sobre a vírgula de Oxford é a mais correta.

E o outro lado merece perder. Tem o pior time. Não há mais espaço para a dialética, para um bate-bola de ideias, uma troca de conceitos no meio de campo, um drible argumental,  uma goleada de contradições.

Não tem VAR, não tem tribunal esportivo, não tem apelação para a FIFA. Nós só queremos ganhar, só a taça importa.

Esta maneira de ver o mundo -o meu time está sempre certo, é o melhor- aparentemente é unânime nos dias de hoje.

Só que toda unanimidade é burra.

Estamos presos nos nossos times. Não conseguimos elogiar, e às vezes, nem reconhecer o valor do outro lado. Sabe o brasileiro incapaz de elogiar um Maradona, o português que acha que Messi não é nem tão bom assim?

Estamos assim em praticamente todas as conversas.

O furacão é um processo natural. O jardim foi plantado. De que lado você vai se posicionar?

A sua escolha revela a sua vida, como ela é.

Este post é o 4 de um exercício chamado #the100DaysProject

Busquei em todos os bolsos de todos os casacos. Fui pro escritório. Voltei. Revirei as roupas sujas, as gavetas, as memórias e as lembranças.

Angustiei-me o dia inteiro por causa de fones de ouvido que custaram vinte euros. Mas eu gosto deles e não queria ter perdido.

Parecia que realmente tinham criado pernas. Aprendido a se camuflar e ido para a guerra.

Escafedeu-se.

Sumiu.

Deu pro mundo.

Existem objetos mínimos que se tornam extremamente importantes. Engraçado como às vezes gastamos rios de dinheiro, tempo e atenção com algo que raramente usamos (barco, casa de praia, jet ski, essas coisas que só dão alegria quando a gente compra ou quando vende), e outras coisas não custam praticamente nada, mas sem elas, o dia instantaneamente perde a graça.

Quarenta minutos no transporte público é um inferno sem fones de ouvido. Oitenta minutos por dia, trezentos dias por mês. E a gente tem a cara de pau de achar que um fone ruim é suficiente.

Travesseiro. Quem economiza com travesseiro pra mim deveria fazer um curso de economia.

Ou de lógica.

Meus fones sumiram e meu dia foi pro saco. Meu humor se acabou, as zoomiões foram um inferno. Tive que fugir.

Até que, do nada, o fone foi encontrado. Pela esposa. Em um lugar que eu tinha olhado pelo menos cinco vezes.

Meu dia, que já estava horrível, fica ainda pior. Mas muito pior mesmo. Porque além de ter perdido o tempo, o humor, o conforto, eu também perdi toda a moral que tinha em casa.

Tive que aguentar o vexame, o ridículo de ser incapaz de encontrar os malditos fones no lugar mais óbvio, onde eu havia guardado, onde sempre estiveram.

Não foram goblins, não foi a Piratinha. Fui eu, o besta, que ignorei o óbvio mais ululante possível.

Mea culpa, mea maxima culpa.

Este post é o 3 de um exercício chamado #the100DaysProject