TechCrunch+

Motional, the Aptiv-Hyundai joint venture developing autonomous vehicle technology, cut its workforce this week — the latest example of layoffs in the AV industry and more broadly in the tech sector.

Employees were told of the layoffs Wednesday, according to sources who asked not to be named because they are not authorized to speak for the company. Motional, which employs more than 1,500 people globally, confirmed the layoffs.

Motional did not confirm the number of employees affected. Sources said dozens of employees were laid off with cuts happening across its operations. Motional has offices in Boston, Las Vegas, Pittsburgh, South Korea, Singapore and several cities in California, including Milpitas and Santa Monica.

“Motional recently announced steps to reallocate resources to areas of the company that will help ensure long-term commercial success,” a company spokesperson said in an emailed statement. “As the organization evolves, we will continue to hire the critical talent needed to develop our technology and meet our commitments to our partners. With our recent partnership announcements and the achievement of technology milestones, we are confident this shift will position us for a strong future.”

The company said it is providing severance packages and outplacement support to help in their transition.

Layoffs have become commonplace in the tech industry this year. The autonomous vehicle sector has been especially hard hit as the high development costs and lack of revenue — never mind profits — has left companies with dwindling cash. Investors, once ready to plow money into this frontier tech, are now looking for safer, nearer term bets.

In November, AV delivery startup Nuro laid off about 300 people, or 20% of its workforce, in an effort to preserve cash amid a stormy economic outlook. Argo AI, the AV startup backed by Ford and Volkswagen, also laid off employees earlier this summer. Several months later, Argo abruptly shut down after its two backers declined to invest more capital into the enterprise and shift its resources to advanced driver assistance systems.

While Motional didn’t provide a reason for its reallocation of resources, other AV companies have cited uncertain economic conditions as well as over-hiring in 2021 as a primary reason. Even AV companies that have yet to lay off employees are feeling pressure to cut costs. Aurora Innovation, a startup-turn-SPAC, presented a swath of cost-cutting and cash-generating options to its board earlier this year, ranging from a hiring freeze and spinning out assets to a small capital raise, going private and even selling itself to high-profile tech companies Apple and Microsoft.

Motional, like many of its remaining AV competitors, has focused its commercialization efforts on launching robotaxi services. The company sees partnerships as a key piece of its go-to-market strategy and has agreements with Lyft, Uber and Via.

Hyundai-backed AV startup Motional cuts workforce by Kirsten Korosec originally published on TechCrunch

Monarch Tractor this morning announced the delivery of its first MK-V unit. The “smart tractor” is electric and what the Bay Area-based company refers to as “driver optional” (terms like “autonomous” and “self-driving” come with their own unique baggage). We’ll just refer to it as a “robot tractor” from here on.

The system was unveiled a bit under two or so years ago. The timing was certainly right. The average age of a U.S. farmer is around 55, and hiring people to help out has become increasingly difficult. Agtech robotics is absolutely a category to be watching closely over the next few years, even amid rough economic headwinds for venture capital.

The first unit is going to wine producer Constellation Brands, which bought a half dozen of the tractors. Additional shipments “to family farms and other large multinational corporations” are also on their way.

“This is a momentous day for Monarch Tractor that has been years in the making,” Monarch co-founder and CEO Praveen Penmetsa said in a release. “Our team has worked tirelessly and relentlessly with a mission to make farming more profitable and sustainable. We’re proud to see our commitment to technological innovation and sustainable solutions culminate in celebrating our first tractor off the production line, which, I might add, is powered by renewable energy.”

The system is powered, in part, by Nvidia’s Jetson platform. That helps manage both driver assistance and the aforementioned “driver-optional modes.” There are several options for that implementation, including a “show mode,” wherein the system follows a worker. That’s managed, in part, by the inclusion of 360-degree camera systems for monitoring its surroundings and implementing various safety features.

Those imaging systems also play a role in another big piece of this. Data collection is arguably the most important role these sorts of systems play. Here that means the ability to monitor larger trends in crop health, yield and growth. These sorts of things are extremely important to farmers when it comes to managing current and future crops.

Monarch has thus far raised $100 million, including a $61 million round last year. It is set to begin manufacturing its systems in Ohio starting next year.

Monarch delivers its first robot tractor by Brian Heater originally published on TechCrunch

Tesla is offering Model 3 and Model Y buyers in the U.S. a $3,750 credit if they have their vehicle delivered in December 2022, according to an update on the company’s inventory page and several posts on Reddit.

Tesla didn’t share the reason for such generosity or publish the news widely — customers received the offer through messages from their dealers. It’s possible the automaker wants owners to take their deliveries before the year is out so it can boost its fourth-quarter sales numbers, which might be dwindling as customers push their delivery times out to 2023 in order to be eligible for the electric vehicle tax credit.

Tesla’s cars haven’t been eligible for an EV tax credit for some time — since the automaker reached the previous cap of 200,000 vehicles sold. But with President Biden’s Inflation Reduction Act (IRA), that cap will be waived by January 1.

Under the new legislation, zero emissions vehicles will be eligible for up to $7,500 in tax credits if automakers can show that their battery components were made or assembled in North America and that a certain percentage of battery critical materials were extracted or processed in countries with which the U.S. has a free trade agreement. If an automaker can only show half, then they’ll only get half of the rebate, which will probably be the case for the next few years, considering most critical materials are still sourced from China.

Which brings us back to the $3,750 discount — the same amount of money a Tesla buyer is expected get back as a rebate next year. But instead of waiting for tax season, Tesla’s offer allows customers to reap the benefits now.

Tesla’s discount also doesn’t discriminate based on income or vehicle manufacturer suggested retail price (MSRP), according to messages received by Tesla customers. Under the IRA’s tax incentive, single tax filers are eligible if their income is below $150,000; heads of households if income is below $225,000; and joint filers’ if income is below $300,000. Additionally new electric cars and SUVs that cost more than $55,000 and $80,000, respectively, don’t qualify for the tax credit.

The Model 3, a compact car, starts at around $47,000 but can easily go over $66,000 depending on model, trim and year. And the 2022 Tesla Model Y starts at $64,990, but a fully loaded performance model can go for more than $80,000.

It’s not common for Tesla to provide discounts — in fact, it’s more like the automaker to increase its price. So the upcoming tax credit might not be the only reason Tesla is dangling discounts. A new report from S&P Global Mobility found that Tesla’s market dominance in the U.S. is waning. The company still dominates the EV sector with its 65% market share, but that’s down from 79% in 2020 and might drop below 20% by 2025. The loss of market share comes as other automakers roll out more affordable EVs.

Tesla offers $3,750 discount for Model 3, Model Y deliveries in December by Rebecca Bellan originally published on TechCrunch

The technological advances happening within the space industry are nothing short of dramatic — from propulsion systems, heavy-lift rockets and increased payload capabilities to building commercial space stations to fill the gap when the International Space Station ceases operations in 2030.

These are thrilling times, and TC Sessions: Space 2022, which takes place on December 6 in Los Angeles, is where you’ll meet the leading founders, investors and makers building the future of space exploration today.

Countdown to launch: Buy a pass now and be in the room to learn, engage, connect and network your way to a stronger business.

Take a deep dive through the impressive agenda to plan your schedule, and be sure to save time to meet and greet the early-stage startups exhibiting equally impressive technology, products and possibilities on the show floor.

Meet the startups exhibiting at the show:

Astrum Drive Technologies
At Astrum Drive Technologies we merge science and technology.

Innovations Health Systems, LLC
Our approach is critical to long-term human health & fitness in MGE

JELLYSPACE
JELLYSPACE removes the entry barriers into the space industry by allowing companies, Startups and organizations to share and access Intellectual Property in safe and secure manner.

Morpheus Space
Morpheus Space is redefining satellite propulsion, serving as the support for those innovators chasing the exciting dream of space exploration.

Nooroot Inc.
Nooroot is a deep-tech company which offers unique data analysis and visualization capabilities inspired by theoretical physics and pure mathematic principles such as the Superstring theory. Our platform technology enables us to unify, analyze, and visualize streaming aeronautical data as well as terrestrial data from any source/sensor in real-time transparently and sustainably.

Plasmos Inc
Reusable last mile delivery for satellites with a unique propulsion system

TRL11, Inc
We make space become more than just mature (TRL9), we make it become mainstream, (TRL11).

Vermeer
Vermeer builds vision based Mixed Reality & AI Enabled Visualization and Command and Control navigation systems for GPS denied and degraded environments. Our solution works on ground vehicles, mobile devices, aircraft, and sUAS.

We’re excited to host some of the most influential leaders, investors and decision-makers in the space industry. You’ll hear from Jory Bell, general partner at Playground Global; Frank Calvelli, assistant secretary of the Air Force for Space Acquisitions; Steve Jurczyk, co-founder and CEO of Quantum Space; Melanie Stricklan, co-founder and CEO at Slingshot Aerospace; and many more.

TC Sessions: Space takes place on December 6 in Los Angeles. Buy your pass today, and then join us to see and learn about the latest tech from established space companies, rising-star founders, and then network for opportunities to build a stronger startup.

Is your company interested in sponsoring or exhibiting at TC Sessions: Space? Contact our sponsorship sales team byfilling out this form.

Meet the early-stage startups exhibiting at TC Sessions: Space by Lauren Simonds originally published on TechCrunch

While artificial intelligence long ago surpassed human capability in chess, and more recently Go — and let us not forget Doom — other more complex board games still present a challenge to computer systems. Until very recently, Stratego and Diplomacy were two of those games, but now AI has become table-flipping good at the former and passably human at the latter.

On the surface, you might think that it’s just because these games require a certain level of long-term planning and strategy. But so do Go and chess, just in a different way.

The crucial difference is actually that Stratego and Diplomacy are games of strategy based on imperfect information. In chess and Go, you can see every piece on the board. Stratego hides the identity of pieces until they are encountered by another piece, and Diplomacy is largely about establishing agreements, alliances and, of course, vendettas that are kept secret but are core to the gameplay. No honest chess game will involve a third party swooping in to protect your opponent’s bishop with a blue rook.

Both games require not raw calculation of paths to victory, but softer skills like guessing what the opponent is thinking, and what they think the computer is thinking, and make moves that accommodate and hopefully upset those assumptions. In other words, it has to bluff and convince another player of something, not just overpower it with the best possible moves.

The Stratego-playing model, from DeepMind, is named DeepNash, after the famous equilibrium. It is focused less on clever moves and more on play that can’t be exploited or predicted. In some cases this can be bold, like one game the team watched against a human player where the AI sacrificed several high-level pieces, leaving it at a material disadvantage — but it was all a calculated risk to bring out the other player’s big guns, so it could strategize around those. (It won.)

DeepNash is good enough that it beat other Stratego systems almost every time, and 84% of the time versus experienced humans. Because the algorithms that work well in Go and chess don’t work well here, they invented a new algorithmic method called Regularised Nash Dynamics — but you’ll have to read the paper if you want to understand it any more deeply than that. In the meantime, here’s an annotated game:

On the Diplomacy side, we have an AI named Cicero (ah, hubris!) from Meta and CSAIL that manages to play the game at a human level — and if that sounds like damning with faint praise, remember Diplomacy is difficult for most humans to play at a human level. The level of scheming, backstabbing, false promises and general Machiavellian antics that people get up to in the game are such that it is banned from many friendly gaming groups. Is a computer really capable of that level of shenanigans?

Seems so, and the advances that make it possible are interesting. After all, the interesting part of Diplomacy isn’t the world map and pieces, which are fairly straightforward to read and evaluate, but the potential for schemes latent in those arrangements. Is Venice being threatened on two fronts, or is it luring the western front into an envelopment through a long contemplated volte-face?

Not only that, but in order to participate in the scheming, one must speak (or chat, online) to other players and convince them of your sincerity and intent. This takes more than CPU cycles!

Here’s how Cicero works:

Using the board state and current dialogue, make an initial prediction of what everyone will do.
Refine that prediction using planning and then use those predictions to form an intent for itself and its partner.
Generate several candidate messages based on the board state, dialogue and its intents.
Filter the candidate message to reduce nonsense, maximize value and ensure consistency with our intents.

Then, plea your case and hope the other player isn’t planning your demise.

When set loose on webDiplomacy.net, Cicero played quite well against its opponents, placing 2nd out of 19 in a league and generally outscoring others.

It’s still very much a work in progress — it can lose track of what it’s said to others, or make other blunders humans probably wouldn’t — but it’s pretty remarkable that it can be competitive at all.

Now AI can outmaneuver you at both Stratego and Diplomacy by Devin Coldewey originally published on TechCrunch

Launch provider Rocket Lab is establishing a subsidiary to handle its sensitive U.S. government business, like launching spysats and experimental military spacecraft. Rocket Lab National Security LLC will handle most of the Defense Department stuff going forward to save the rest of the company a bit of grief.

The main arm of Rocket Lab, or perhaps it would be more appropriate to call it the body, has done plenty of business with the U.S. government already, putting tests from DARPA and the National Reconnaissance Office into orbit. It’s kind of inevitable if you’re a launch provider — governments in general are big customers, and the U.S. in particular.

But wow, you want to talk about picky? Try getting a contract to launch a top secret satellite that costs $200 million! There are all kinds of hoops you have to jump through. But then they want to launch right away, price is no object — as if you’ve just got rockets lying around!

Of course in Rocket Lab’s case that might well be true, but the company clearly decided that it would be smart to contain the “bureaucratic requirements” and other red tape, clearances, etc to a specialized subsidiary that can work more closely with its national security clients and partners. It’s likely there was already a considerable firewall within the company, since commercial activity is in many ways fundamentally different from government contracts. Now it’s truly its own business unit (or perhaps that is not the correct term of art, but you know what I mean).

“Top of the list for national security is reliability and responsiveness, something we’ve delivered on across multiple missions already. With Rocket Lab National Security we’re building on this strong heritage to deliver tailored capabilities that evolve as the nation’s needs do,” said Brian Rogers, leader of the company’s government launch services department, in a press release.

Less high-touch missions for civilian agencies like NASA and NOAA would likely use the “vanilla” Rocket Lab’s services, and indeed defense projects that don’t come with too many strings attached can probably save a buck or two by going that way as well. But compared with launch costs ten years ago it’s all a rounding error.

The launch company’s work designing spacecraft and on more complex missions like CAPSTONE remain part of the main company as well. Until the military starts launching space lasers to cislunar space, anyway.

Rocket Lab carves off defense-focused subsidiary for national security customers by Devin Coldewey originally published on TechCrunch

In the older days of the web, an app called Trillian emerged to solve the problem of allowing users to connect to their friends across multiple messaging services, like AIM, Jabber, Yahoo! Messenger, IRC and others.

Lately, I’ve been longing for a Trillian-like service that would allow me to keep up with Twitter’s fleeing user base, the way that Trillian once connected users to multiple IM services.

I’m not sure that would be a good thing, though.

In the days since the Elon Musk Twitter takeover, a good chunk of Twitter’s network has exited to other apps. That leaves Twitter users who are trying to stay connected with friends with a lot of open browser tabs and new apps stuffed into iPhone folders.

The problem is, the market hasn’t yet picked a winner for “new Twitter” — and it’s possible it never will.

It’s a tough problem to solve. Reproducing Twitter’s magic isn’t as simple as re-creating a microblogging tool with similar functionality — a heart, a retweet, an @ reply and threads. Twitter’s biggest draw was its community, not its feature set. And it’s fair to say that network has been diminished in recent days — despite Musk’s tweets to the contrary. Even if Twitter is gaining users, as he claims, it’s losing others at the same time.

If Musk succeeds, Twitter’s sheer numbers may grow, but it won’t be the same Twitter it was before. Much like the company itself, when the original people are gone, so is the vibe, the spark, the sense of belonging. MySpace, after all, is still around. But it’s not MySpace anymore.

Or, in other words: though Twitter lives on, in some ways, it’s already gone.

You’ve probably felt it, too. Every day since the Musk takeover, much of my Twitter engagement now comes from crypto trolls who are tagging me into threads and bizarro, hashtagged praise for Musk from people’s accounts that aren’t quite discernible from bots. Meanwhile, real conversations are happening elsewhere — and with increasing frequency.

Twitter, meanwhile, seems to be doubling down on its reward-and-punishment system for only the most performative, most snarky, most zingy of posters, led by twit-in-chief Musk. Ultimately, it seems, Twitter may live on as an entertainment network, if not a community, as a result.

In the meantime, many of Twitter’s OG users have fled to different parts of the web.

Those with more technical expertise may have managed to navigate the complicated onboarding process to find a home on Mastodon, but it’s not a Twitter clone despite the initial similarities. Universal search doesn’t exist. Direct messages work differently. Quote tweets aren’t a thing. The mobile app is slow. Adding friends is wonky. And there are so many rules. (You’re even supposed to add content warnings to talk about Twitter, it’s so despised over there! And yet Twitter users’ collective sadness over its seeming downfall is all we want to talk about now. )

Elsewhere, some of Twitter’s media crowd have been trying out Post.News, which is being advised by tech journo Kara Swisher. But it’s not launched publicly yet and there’s no mobile app. Plus, the network itself, with its journalist base, is sort of …serious and boring. (Sorry, but it wasn’t media Twitter that made Twitter fun!)

Some young people left Twitter for Hive, whose 3-person team can’t keep the darned thing functioning. Currently, it’s offline as the team addresses significant security concerns. The company’s lack of transparency around staffing, funding, security, privacy and other key aspects of its business — not to mention its tendency to call everyone “besties” — doesn’t inspire much trust, either.

Then there are the numerous alt Twitters: Cohost, CounterSocial, Tribel, Koo, and so on.

Oh, and I guess we’re using Tumblr again, too? Which may one day connect to Mastodon folks? Substack wants to be a Twitter now? And Discord?

But wait! There’s more. Twitter co-founder Jack Dorsey is building his own decentralized Twitter called BlueSky, which is both a protocol and an app. But BlueSky says it won’t use ActivityPub — which means it won’t be able to talk to Mastodon. Sigh.

Even as all the alt Twitter sampling continues, many Twitter users haven’t actually fully left — they’ve just reduced their postings or are cross-posting. They’ve got one foot out of the door but keep looking back over their shoulder, as if asking themselves, should I really leave? Is now the time? (Points at self). After all, it seems there’s still too much going on over on Twitter to truly sign out forever.

I’m tired. This problem is in need of a solution. Or at least a decision. Can we convene a vote, please?

Eventually — hopefully — the issue will resolve itself.

Maybe we’ll collectively decide that we actually don’t choose which companies to use based on our morals and values, as it turns out. (Or else our local mom-and-pop stores would have our business, not Amazon Prime!) Maybe, with heads hung low, we’ll go back to Twitter and pretend we never left.

Or maybe we’ll all finally gravitate towards one network over the others, which will then gain enough traction to consider itself the real “Twitter 2.0.”

Or maybe we’ll decide we don’t need a Twitter 2.0, and stay dispersed across smaller networks. Maybe that’s the right move. Maybe it’s time to retire the concept of the global town square, the dunk factory, and the algorithm.

Maybe we should find a group of people we actually want to talk to — not perform for — and leave Twitter behind.

What’s a Twitter user to do? by Sarah Perez originally published on TechCrunch

Released two years ago, OpenAI’s remarkably capable, if flawed, GPT-3 was perhaps the first to demonstrate that AI can write convincingly — if not perfectly — like a human. The successor to GPT-3, most likely called GPT-4, is expected to be unveiled in the near future, perhaps as soon as 2023. But in the meantime, OpenAI has quietly rolled out a series of AI models based on “GPT-3.5,” a previously-unannounced, improved version of GPT-3.

GPT-3.5 broke cover on Wednesday with ChatGPT, a fine-tuned version of GPT-3.5 that’s essentially a general-purpose chatbot. Debuted in a public demo yesterday afternoon, ChatGPT can engage with a range of topics, including programming, TV scripts and scientific concepts.

According to OpenAI, GPT-3.5 was trained on a blend of text and code published prior to Q4 2021. Like GPT-3 and other text-generating AI, GPT-3.5 learned the relationships between sentences, words and parts of words by ingesting huge amounts of content from the web, including hundreds of thousands of Wikipedia entries, social media posts and news articles.

Rather than release the fully trained GPT-3.5, OpenAI used it to create several systems fine-tuned for specific tasks — each available through the OpenAI API. One — text-davinci-003 — can handle more complex instructions than models built on GPT-3, according to the lab, and is measurably better at both long-form and “high-quality” writing.

According to OpenAI data scientist Jan Leike, text-davinci-003 is similar but not identical to InstructGPT, a family of GPT-3-based models released by OpenAI earlier this year that are less likely to generate problematic (e.g., toxic and highly biased) text while more closely aligning with a user’s intent. Text-davinci-003 — and by extension GPT-3.5 — “scores higher on human preference ratings” while suffering from “less severe” limitations, Leike said in a tweet.

That appears to be the case anecdotally. Data scientists at Pepper Content, a content marketing platform, report that text-davinci-003 “performs better in understanding the ‘context’ behind a request and then using that to produce better content” while “hallucinating” less than GPT-3-based models. (Where it concerns text-generating AI, hallucination refers to an AI writing inconsistent, factually incorrect statements.)

In a test on OpenAI’s Playground website, which provides a UI frontend for the models, the Pepper Content team fed several prompts to text-davinci-003 and a model based on GPT-3 (text-davinci-002). Given “What is the philosophy behind WeWork?,” the GPT-3.5-based text-davinci-003 generated this:

WeWork is a global workspace provider that believes people are the most important asset in any organization. The philosophy of WeWork is to create a collaborative environment that enables people to work together in a flexible and efficient way. The company encourages collaboration and productivity, while providing a comfortable and inspiring space.

WeWork is also committed to being a socially responsible organization, by finding ways to reduce its environmental impact, by providing meaningful work experiences, and by promoting diversity and inclusion. WeWork also strives to create meaningful experiences for its members, through its unique community-based programming, events and activities. The company believes that when people work together in an inspiring and collaborative environment, they can achieve more and create meaningful change.

It’s not perfect — note the excess commas and repetitiveness. But the copy’s certainly more engaging than what the GPT-3-based text-davinci-002 produced:

At WeWork, we believe in openness, collaboration, and community. We have reimagined what a workspace can be by bringing together a global community of creators, entrepreneurs, and startups — anyone looking to build something meaningful and transform the world. We are proud to be your neighbour, coworker, and friend.

GPT-3.5 is also better at generating blog posts, it seems. Here’s what the Pepper Content team got when they prompted text-davinci-003 to write a post about picking a sofa:

If you are looking to give your green living room a fresh look, adding the the best sofa for the space is a great way to start. However, there is more to choosing the perfect sofa than just finding a piece that looks good. You want to find a piece that is comfortable and fits your green living room décor.

In this blog, we will discuss how to pick sofa far your green living room. We will look at the different types of sofas available, the design factors to consider when making your selection, and how to choose a sofa that fits your style and budget. We will also provide tips on how to maintain and care for your new sofa. By the end of this blog, you will have all the information you need to pick the perfect for your green living room.

Again, it isn’t foolproof. GPT-3.5 oddly added the bit about a “green living room.” But also again, GPT-3 is more basic and less grammatical in its generation:

Sofa is one of the most basic requirements in a living room. It’s not just a piece of furniture but an important part of the décor of your living room. So, what should be the criteria while picking a sofa? If you are wondering about this then stay with me as I discuss the different aspects of the sofa would help you in picking the best one for yourself.

Experiments beyond Pepper Content’s suggest that GPT-3.5 tends to be much more sophisticated and thorough in its responses than GPT-3.

For example, when YouTube channel All About AI prompted text-davinci-003 to write a history about AI, the model’s output mentioned key luminaries in the field, including Alan Turing and Arthur Samuelson, while text-davinci-002”s did not. All About AI also found that text-davinci-003 tends to have a more nuanced understanding of instructions, for instance providing details such as a title, description, outline, introduction and recap when asked to create a video script.

A hallmark feature of text-davinci-003/GPT-3.5’s generations is wordiness, as it turns out. In an analysis, scientists at startup Scale AI found text-davinci-003/GPT-3.5 generates outputs roughly 65% longer than text-davinci-002/GPT-3 under identical prompts.

Perhaps less useful but nonetheless entertaining, text-davinci-003/GPT-3.5 is better at composing songs, limericks and rhyming poetry than its predecessor. Ars Technica reports that commenters on Y Combinator’s Hacker News forum used text-davinci-003 to write a poem explaining Albert Einstein’s theory of relativity and then re-write the poem in the style of John Keats. See:

If you want to understand Einstein’s thought
It’s not that hard if you give it a shot
General Relativity is the name of the game
Where space and time cannot remain the same
Mass affects the curvature of space
Which affects the flow of time’s race
An object’s motion will be affected
By the distortion that is detected
The closer you are to a large mass
The slower time will seem to pass
The farther away you may be
Time will speed up for you to see

The Scale AI team even found that Text-davinci-003/GPT-3.5 has a notion of meters like iambic pentameter. See:

O gentle steeds, that bear me swift and

sure

Through fields of green and pathways so

obscure,

My heart doth swell with pride to be with

you

As on we ride the world a-fresh to view

The wind doth whistle through our hair so

free

And stirs a passion deep inside of me.

My soul doth lift, my spirits soar on high,

To ride with you, my truest friend, am I

Your strength and grace, your courage and

your fire,

Inspire us both to go beyond our sire.

No earthly bonds can hold us, only fate,

To gallop on, our wond’rous course create

Relatedly, text-davinci-003/GPT-3.5 is wittier — at least subjectively. Asking text-davinci-002/GPT-3 to “tell a joke” usually yields this:

Why did the chicken cross the road? To get to the other side.

Text-davinci-003/GPT-3.5 has cleverer responses:

Q: What did the fish say when it hit the wall? A: Dam!

Q: What did one ocean say to the other ocean? A: Nothing, they just waved.

Scale AI had it explain Python code as Eminem, a feat which text-davinci-002/GPT-3 couldn’t accomplish:

Yo, so I’m loopin’ through this list

With each item that I find

I’m gonna print out every letter in each one

of them

Dog, Cat, Banana, Apple, I’m gonna get’em

all with this rhyme

So why is GPT-3.5 so much better than GPT-3 in these particular areas? We can’t know the exact answer without additional details from OpenAI, which aren’t forthcoming; an OpenAI spokesperson declined our request for comment. But it’s safe to assume that GPT-3.5’s training approach had something to do with it. Like InstructGPT, GPT-3.5 was trained with the help of human trainers who ranked and rated the way early versions of the model responded to prompts. This information was then fed back into the system, which tuned its answers to match the trainers’ preferences.

Of course, it doesn’t make GPT-3.5 immune to the pitfalls to which all language models eventually succumb. Because GPT-3.5 merely relies on statistical regularities in its training data rather than a human-like understanding of the world, it’s still prone to, in Leike’s words, “mak[ing] stuff up a bunch.” It also has limited knowledge of the world after 2021 because the training data is more sparse after that year. And its safeguards against toxic output can be straightforwardly circumvented.

Still, GPT-3.5 and its derivative models demonstrate that GPT-4 — whenever it arrives — won’t necessarily need a huge number of parameters to best the most capable text-generating systems today. (Parameters are the parts of the model learned from historical training data and essentially define the skill of the model on a problem.) While some have predicted that GPT-4 will contain over 100 trillion parameters — nearly 600 times as many as GPT-3 — others argue that emerging techniques in language processing, like those seen in GPT-3.5 and InstructGPT, will make that enormous jump unnecessary.

One of those techniques could involve browsing the web for greater context, a la Meta’s ill-fated BlenderBot 3.0 chatbot. John Shulman, a research scientist and co-founder of OpenAI, told MIT Tech Review that OpenAI is continuing work on a language model it announced late last year, WebGPT, that can go and look up information on the web (via Bing) and give sources for its answers. At least one Twitter user appears to have found evidence of the feature undergoing testing for ChatGPT.

OpenAI has another reason to pursue lower-parameter models as it continues to evolve GPT-3: huge costs. A 2020studyfromAI21 Labs pegged the expenses for developing a text-generating model with only 1.5 billion parameters at as much as $1.6 million. OpenAI has raised over $1 billion to date from Microsoft and other backers, and it’s reportedly in talks to raise more. But all investors, no matter how big, expect to see returns eventually.

While anticipation builds for GPT-4, OpenAI quietly releases GPT-3.5 by Kyle Wiggers originally published on TechCrunch

It might seem like Lensa AI sprung up over night when suddenly, your friends are all posting artistic renditions of themselves that they generated on the app. But while the Lensa AI app itself has been around since 2018, its viral “magic avatars” feature launched in late November, boosting the app to the #1 spot on the iOS App Store‘s competitive “Photo & Video” charts. For comparison, YouTube is #3 and Instagram is #4 on the charts at the time of publication.

Lensa AI works by inviting users to upload 10-20 photos of themselves. Using the open source Stable Diffusion model, the app processes your photos to generate avatars of you that look like they were created by a digital artist.

When you download Lensa AI, you’re immediately greeted with a pop-up inviting you to join a seven-day free trial to use their AI editing tools — if you don’t cancel in time, you’ll be charged $39.99 for unlimited use of the app for a year. You can bypass this screen without committing to the trial, but the free version of the app is very limited and doesn’t include the viral magic avatar feature. Even if you don’t subscribe to the unlimited plan, you’ll be hit with another in-app purchase screen if you try to make your very own magic avatar. At the lowest price tier, you’ll have to pay $3.99 for 50 unique avatars (5 variations of 10 different styles). But hey, at least that’s cheaper than a blue check!

According to app analytics firm SensorTower, Lensa AI has amassed about 22.2 million worldwide downloads and almost $29 million in consumer spending since its launch in 2018. In the last month, the app has seen a significant spike with the release of magic avatars. In November, the app was downloaded 1.6 million times, up 631% from 219,000 downloads in October. The U.S. is Lensa AI’s largest market, generating 58% of consumer spending, but the app is especially popular in Brazil right now. Of all November downloads, 31% were from Brazil, where installs climbed 24,450% month over month.

Viral photo editing apps have a shady history, as some apps have been found to be vectors for malware. In other cases, users have worried about what happens to the photos they upload into these apps. These concerns came up around Russia-based AI editor FaceApp, which later made a statement that it might store updated photos in the cloud for “performance and traffic reasons,” but that most images are deleted within 48 hours.

Prisma Labs, the team behind Lensa AI, told TechCrunch that it uses AWS cloud services to process user’s photos. As soon as an AI model is trained on a user’s photos, the images are immediately deleted.

Lensa AI climbs the App Store charts as its ‘magic avatars’ go viral by Amanda Silberling originally published on TechCrunch

Apple iPhone users with iOS 16.1 software can now share car keys in their Wallet with non-iPhone users, starting with Google Pixel devices. In the future, that capability will extend to other devices with Android 12+, according to Apple.

Apple says it has been working with the Internet Engineering Task Force and other industry players to establish a standard for sharing digital keys across platforms. Keys can be shared via email, text message and WhatsApp.

When Apple initially launched digital car keys in 2020, iPhone users could share their keys through iMessage, Apple’s instant messaging service.

Of course, key sharing only works on cars that are compatible with digital car keys. Apple didn’t share a list of compatible cars, and instead directed users to contact their car manufacturer or dealership for an answer. To our knowledge, there are several 2021 and 2022 models of BMW that can be unlocked and used via digital car keys on iPhones, Google Pixels, Samsung Galaxy devices. Both Apple’s and Google’s digital car keys also recently arrived on the 2022 Kia Niro, the 2022 Genesis G90 and the all-electric 2022 Genesis GV60.

Apple’s digital car keys can usually be added to the Apple Watch (Series 5 or later or the Apple Watch SE) with the latest version of watchOS, according to a post from Apple. It’s not clear if this cross-platform capability will be available for Apple Watch users just yet.

Apple’s digital car keys are one of the company’s many features that allow users to carry less stuff around. Perhaps one of the company’s most popular examples of this is Apple Pay, which syncs a user’s credit, debit or prepaid card to their phone. In addition, iPhone users in certain states can now carry their driver’s license or state ID in their Wallet, and in select Hyatt hotels, users can unlock their rooms with a digitally stored key.

iPhone users can share car keys in Wallet with non-iPhone users by Rebecca Bellan originally published on TechCrunch