The Neural Net

The Rise of ChatGPT:

In 2018, many writers began to use AI in their novels

It helped them write quickly, but it also raised complex aesthetic and ethical questions

With the debut of chatGPT, these questions have become more urgent and mainstream

Jennifer Lepp, who writes under the pen name Leanne Leeds, is still using Sudowrite, but has begun incorporating some of the more recent tools into her fiction

She hopes that it will help her write two books at the same time, but hasn't been able to do that yet

Everyone is now talking about using chatgpt on the peripherals, but there seems to be a moral chasm between the two

"It does blurbs really well, and I hate doing blurbs, so I'm going to use it for blurbs." - "Well, I'll use it to help me tighten up my plot."

What do you think the line is that people are drawing?

It's a concern of plagiarism.

Everybody knows that they copy stuff with permission and without permission. It's just much more obvious in the artist community.

Do you see a line between using AI for something like a description and using it to mimic another author's voice? That I won't do.

ChatGPT-generated plots for books 2, 3, 4, 5, 6, and 7

All of those murder mysteries were ChatGPT-generated, though I edited some of them.

The impressive thing about it is that if I tell it that it's a cozy mystery and it needs to be humorous, it seems to understand what I'm asking for.

You mentioned over email that you were using AI for book covers. How did you come up with the idea for the Lykoi cat cover?

Wouldn’t have been able to find a photographer to do the shoot, and it would have cost a lot to get the image and cover that she needed.

Instead, she used AI to draw inspiration for the cat, which is a crossbreed of a hairless and a werewolf hybrid.

How do you see these tools and the way writers use them evolving?

I'm really just stuck in the middle, wondering which way it's going to go. I definitely don't want to encourage people who aren't comfortable using it to use it.

ChatGPT shocked the hell out of me. I had thought, well, it'll take three or four years, and it'll get better. Then came ChatGTT, and oh my god, that's so much better!

Art platform ArtStation is removing images protesting AI-generated art from its homepage.

Claiming that the content violates its Terms of Service.

Protestors are concerned that the AI art is derivative of the labor of human artists and often uses their work without attribution or compensation.

Artstation is used by creatives to display industry portfolios and find paid work, so it is not surprising that the community has objected to the apparent acceptance of AI art on the platform.

The decision to remove AI protest imagery from ArtStation has, unsurprisingly, gone down poorly with the site's community, with many users now declaring their intention to abandon the platform altogether.

This isn't the first time Artstation has come under fire for censoring artwork, having been accused of removing anti-war and pro-Ukraine content earlier this year.

AI chatbots are still too immature to put in front of users.

Concerns include bias, toxicity, and their propensity for simply making information up.

Google has developed a number of large AI language models (LLMs) equal in capability to OpenAI’s ChatGPT, including BERT, MUM, and LaMDA, which focus on parsing users’ queries to better understand their intent.

A friend of mine texted me earlier this week to ask what I thought of ChatGPT. I wasn't surprised he was curious. He knows I write about AI and is the sort of guy who keeps up with whatever's trending online. We chatted a bit, and I asked him: "and what do you think of Chat

"Capability Overhang" refers to the hidden capacities of AI: skills and aptitudes latent within systems that researchers haven't even begun to investigate yet.

Today's models are far more capable than we think, and our techniques available for exploring them are very juvenile

Capability overhang is a technical term, but it also perfectly describes what's happening right now as AI enters the public domain

A glut of new apps and programs have suddenly made these skills available to a general audience, and in 2023, as we continue scaling this new territory, things will start changing fast

Stack Overflow has temporarily banned users from sharing responses generated by AI chatbot ChatGPT.

The bot's interface encourages people to ask questions and in return offers impressive and fluid results across a range of queries; from generating poems, songs, and TV scripts, to answering trivia questions and writing and debugging lines of code.

However, some users have noted its persistent tendency to generate plausible but false responses.

Google has licensed its AI research model for breast cancer screening to medical technology company iCAD

The hope is that AI might become a tool to help radiologists and their patients

Commercial deployment, however, still depends on how successful continued research and testing are

Google aims to eventually deploy the technology in real-world clinical settings - targeting a "2024 release"

Generally, medical experts are approaching AI with caution

Relying too heavily on AI without bringing doctors in to assess all the nuances of a patient's health could heighten the risk of overdiagnosis

Changes to the AI text-to-image model make it harder for users to mimic specific artists' styles or generate NSFW output, but offer other functional improvements

Stability AI, the company that funds and disseminates the software, announced Stable Diffusion Version 2 early this morning European time.

The update re-engineers key components of the model and improves certain features like upscaling (the ability to increase the resolution of images) and in-painting (context-aware editing).

However, the changes also make it less able to generate certain types of images that have attracted both controversy and criticism including nude and pornographic output, photorealistic pictures of celebrities, and images that mimic the artwork of specific artists.

A team of machine learning engineers from Facebook's parent company Meta has unveiled a new system called Make-A-Video that allows users to type in a rough description of a scene, and it will generate a short video matching their text. The videos are clearly artificial, with blurred subjects and distorted animation, but represent a significant development in the field of AI content generation.

The best way to judge the model's performance is to watch its output

Each of the videos below was generated by Make-A-Video and captioned with the prompt used to generate the video.

However, note that each video was provided to The Verge by Meta, which is not currently allowing anyone access to the model.

What are the limitations?

The model has many technical limitations beyond blurry footage and disjointed animation.

Like all AI models trained on data scraped from the web, Make-A-Video has "learned and likely exaggerated social biases, including harmful ones."

It is impossible to say what biases Meta's model has learned without open access.

Onstage at the launch of Amazon's Alexa Prize, a multimillion-dollar competition to build AI that can chat like a human, the winners of last year's challenge delivered a friendly warning to 2018's hopefuls: your bot will mess up, it will say something offensive, and it will be taken offline.

Teaching a machine how to have a real conversation is one of AI's hardest challenges

There's just so much that computers don't know about the world, and there's no easy way for us to teach them

The Alexa Prize is a competition that will help shape the future of voice-based computing for years to come

Amazon isn't asking much: just create a chatbot using Alexa that can talk to a human for 20 minutes without messing up

Each of the eight teams will be building their chatbots using Amazon's resources: speech recognition tools from Alexa, free computing power from Amazon Web Services, and stacks of training data from tens of millions of Alexa users

In last year's inaugural competition, the University of Washington's chatbot was able to make successful conversation for just over 10 minutes on average, which still leaves the grand prize up for grabs this year

Teams get to compete, and Amazon gets to pick talent

It also gets the opportunity to hire any particularly promising researchers and essentially crowdsource future technological paths for its AI assistant

Good conversational AI requires personality as well as patter

A lot of automation starts as data created by human labor

We may be using machines to build conversational bots, but it all starts as human chatter

For the team from Brigham Young University in Utah, finding a source of training data was easy: they turned to their fellow students

To collect data, the group behind the university’s Eve chatbot set up the Chit-Chat Challenge

Chats could be about anything, but couldn't include personally identifying information or topics that wouldn't appeal to people outside the university

BYU is a religious university that is owned and run by the Mormon Church, so the students share a worldview which helps create a more cohesive personality for the bot

To train machines to read words, ask them to read Wikipedia

A neural network scans the text in small windows, centering on one word at a time, but also glimpsing in its peripheral vision the three or four words on either side of it

Learns to predict what words tend to appear alongside one another, and turns this data into a vector representation

The position of the vector is arbitrary, and by itself doesn't capture any meaning related to the word.

But the relationship of one vector to another does, and you can infer properties about words based on their properties

The Future

With the ability to turn sentences into vector representations, the BYU team hopes they can provide a better feedback loop for the replies generated by their AI system

This, they say, will allow the system to teach itself without human intervention, which means quicker training and improved results.

Building AI systems that interact with people takes something closer to artistry.

You have to have an instinct for what works and the patience to slog through what doesn't. You need to strike a balance with the user: give them what they want but not too much of it.

The solution: acknowledge what the user has said, then change the topic and offer up some new direction

It worked.

Voice-based computing is the future

Amazon believes ambient computing is here to stay

Unlike Google Assistant and Apple’s Siri, Alexa was never developed with mobile phones in mind

It's hardware-agnostic

Alexa is a platform as much as a product

The team responsible for Alexa is also building tools so other companies can use Alexa

The Future

In a weird way, Limp gets to look into the future

An AI that can talk like a human might become our most addictive gadget yet

If we could genuinely talk to our devices - not just command but converse with them - then our relationship with the digital world would be upended

The techniques being developed by the teams at this year's Alexa Prize are ingenious and worthy of praise, but chatbots still have a long way to go before they match humanity in its gift of the gab

Waymo, the self-driving unit of Alphabet, is the only company in the world to have fully driverless vehicles on public roads today.

The company asserts that its cars have the most advanced brains on the road today thanks to a head start in AI investment, strategic acquisitions by sister company Google, and a close working relationship with the tech giant's in-house team of AI researchers

Waymo's engineers are modeling how cars recognize objects in the road, how human behavior affects how cars should behave, and they're using deep learning to interpret, predict, and respond to data accumulated from 6 million miles driven on public and 5 billion driven in simulation.

In March, a 49-year-old woman was struck and killed by a self-driving Uber vehicle while crossing the street in Tempe, Arizona.

A few weeks later, the owner of a Tesla Model X died in a gruesome crash while using Autopilot, the automaker’s semi-autonomous driver assist system

Meanwhile, the public is growing increasingly skeptical of the safety of driverless vehicles

In the midst of all this uncertainty, Waymo invited me out to its headquarters in Mountain View, California, for a series of in-depth interviews with the company's top humans in artificial minds.

ImageNet

ImageNet started as a poster from Princeton University researchers, displayed at a 2009 conference on computer vision and pattern recognition in Florida

It grew into an image dataset, then a competition to see who could identify the most images with lowest error rate

Around 2011, the error rate was about 25 percent, meaning one in four images were being identified incorrectly by the teams' algorithms

A big shift was going from neural nets that were quite shallow (two or three layers) to the deep nets (double-digit layers)

The biggest breakthrough was in 2012, when AI researcher Geoffrey Hinton and his two graduate students, Ilya Sutskever and Alex Krizhevsky, showed a new way to attack the problem: a deep convolutional neural network to the ImageNet Challenge that could detect pictures of everyday objects

This led to Google acquiring Hinton’s company DNNresearch

Problems emerged almost immediately: the new system was making too many errors, mislabeling cars, traffic signals, and pedestrians

Google began applying the technique to other parts of the project including prediction and planning

Other car and tech companies have already caught on to the importance of machine learning, and Waymo's data may be too specific to extrapolate to a global scale.

AI and machine learning are essential to self-driving cars, but how long can they last?

A natural consequence of improvements in AI is that big head-starts like Waymo’s are less significant as they have been

Each additional mile that Waymo accrues needs to be interesting for it to be relevant to the process of training the neural networks

Dolgov is sitting in one of X's conference rooms, whiteboard marker in hand, MacBook Pro splayed before him, asking me to describe to him the difference between Garfield and Odie.

Before I can stammer out a reply, Dolgov keeps going: "If I give you a picture, and I ask you "is it a cat or a dog," you will know very quickly."

This type of question is well-suited for deep learning algorithms

It's one thing to come up with a bunch of basic rules and parameters, but teaching a computer to distinguish between different types of traffic signs is much easier

Waymo uses an automated process and human labelers to train its neural nets

After they've been trained, these giant datasets also need to be pruned and shrunk so they can be deployed in the real world in Waymo's vehicles

This process, akin to compressing a digital image, is key building the infrastructure to scale to a global system.

The Future of AI

The future of AI at Waymo isn't sentient vehicles. It's in cutting-edge research like automated machine learning, in which the process of building machine learning models is automated.

Waymo uses Google's data centers to train its neural nets, using a high-powered cloud computing hardware system called "tensor processing units."