Welcome to the TechMeme right home for Tuesday, November 7th, 2023. I'm Brian McCulloch today. Open AI wants to open an app store where everyone can build their own GPs. There's also a new GPs4 that's gotten significantly cheaper. Did crews know its cars were underperforming even before the DMV shut them down? We work as officially dead, but are we seeing grain shoots? Evidence that the tech recession might be ending. Here's what you missed today in the world of tech.
We've discussed how this AI moment is not even a year old yet. ChatGPT was only released November 30th of last year, and I went into the RSS feed yesterday and found that the very first mention of ChatGPT on the show only came on December 5th. So not even a year in, and it's amazing, but OpenAI's developer day is now as important to our purposes. As an Apple event, I think there were two big stories yesterday, so we'll break them up into two segments.
First, OpenAI said ChatGPT has 100 million weekly users, and that more than two million developers are currently building on the company's API, including at over 92% of Fortune 500 companies, and also you know at little developers like resumewriting.com. Well, the first big news was the debut of GPT-4 Turbo with a 128,000 token context window. Also an April 2023 knowledge cutoff, a JSON mode, new modalities in the API, and 2X to 3X cheaper tokens.
Quoting tech crunch, GPT-4 Turbo comes in two versions, one that's strictly text analyzing and a second version that understands the context of both text and images. The text analyzing model is available in preview via an API starting today, and OpenAI says it plans to make both generally available in the coming weeks.
They're priced at 1 penny per thousand input tokens, around 750 words, where tokens represent bits of raw text, e.g., the word fantastic, split into fantastic and tick, and 3 cents per thousand output tokens. Input tokens are tokens fed into the model, while output tokens are tokens that the model generates based on the input tokens.
The pricing of the image processing GPT-4 Turbo will depend on the image size, for example, passing an image with 1080x1080 pixels to GPT-4 Turbo will cost 0.00765, OpenAI says. We optimize performance, so we're able to offer GPT-4 Turbo at 3X cheaper price for input tokens, and 2X cheaper price for output tokens compared to GPT-4. OpenAI writes in a blog post shared with tech crunch this morning.
GPT-4 Turbo boasts several improvements over GPT-4, one being a more recent knowledge base to draw on when responding to requests. Like all language models, GPT-4 Turbo is essentially a statistical tool to predict words. Fed an enormous number of examples, mostly from the web, GPT-4 Turbo, learned how likely words are to occur based on patterns, including the semantic context of surrounding text.
For example, given a typical email ending in the fragment looking forward, GPT-4 Turbo might complete it with two hearing back. GPT-4 was trained on web data up to September of 2021, but GPT-4 Turbo's knowledge cutoff is April 2023. That should mean questions about recent events, at least events that happen prior to the new cutoff date, will yield more accurate answers. GPT-4 Turbo also has an expanded context window.
Context window measured in tokens refers to the text the model considers before generating any additional text. Models with small context windows tend to forget the content of even very recent conversations, leading them to veer off topic often in problematic ways. GPT-4 Turbo offers a 128,000 token context window, four times the size of GPT-4's, and the largest context window of any commercially available model, surpassing even Anthropics Claw II.
Claw II supports up to 100,000 tokens, Anthropic Claims to be experimenting with a 200,000 token context window, but has yet to publicly release it. Indeed, 128,000 tokens translates to around 100,000 words or 300 pages, which for references around the length of weathering heights, gullivers, travels, and Harry Potter and the prisoner of Asgaban.
And GPT-4 Turbo supports a new JSON mode, which ensures that the model responds with valid JSON, the open standard file format and data interchange format. That's useful in web apps, that transmit data, like those that send data from a server to a client, so it can be displayed on a web page, OpenAI says.
Other related new parameters will allow developers to make the model return consistent completions more of the time, and for more niche applications, log probabilities for the most likely output tokens generated by GPT-4 Turbo. But possibly bigger news, OpenAI also introduced GPT's custom versions of chat GPT, that chat GPT plus and enterprise users, can create without any coding and monetize in its upcoming GPT store.
In other words, OpenAI got an app store, but instead of building apps, they want you to build GPT's. Like in theory, I could train a GPT on all of the scripts I produced for the show over the last six years and create a GPT, which would do I don't really know. But you can see that they're obviously reaching for a sort of iPhone moment here, quoting the verge. In the coming weeks, these AI agents, which OpenAI is calling GPT's, will be accessible through the GPT store.
Details on how the store will look and work are scarce for now, though OpenAI is promising to eventually pay creators an unspecified amount based on how much their GPT's are used. GPT's will be available to paying GPT plus subscribers and OpenAI enterprise customers who can make internal only GPT's for their employees. During a recent demo, I received of OpenAI's GPT platform, a bot called Creative Writing Coach, critiqued an uploaded PDF of a writing sample.
Over a period of about two minutes, I watched other GPT be spun up to help attendees navigate DevDay. The platform auto-named the bot event navigator, generated a profile picture for it using.le and ingested a PDF attachment with the event's schedule to inform its answers. OpenAI's interface lets you guide how you want a GPT to interact with people before you publish. The DevDay event navigator, Agenai saw during my demo was instructed to be helpful and concise and to avoid scheduling conflicts.
OpenAI auto-generated several conversation starter prompts, such as what's the first session today? Each GPT can be granted access to web browsing, dolly, and OpenAI's code interpreter tool for writing and executing software. There's also a knowledge section in the builder interface for uploading custom data like the DevDay event schedule.
With another feature called actions, OpenAI is letting GPT's hook into external services for accessing data like emails, databases, and more, starting with Canva and Zapier. The introduction of custom GPT's means that OpenAI is now competing with other AI bot platforms like Character AI and Meta, which recently introduced a slew of its own AI personas in WhatsApp, Instagram, and Messenger.
OpenAI is positioning its platform as being more utility-focused than its competitors rather than emphasizing bots that act like people. Though it isn't against people building GPT's with human-like personas. The creators of GPT's won't be able to view the chats people are having with them, and it's unclear what high-level usage data they'll get access to. OpenAI says it will be monitoring activity to block things like fraud, hate speech, and adult themes.
When the GPT store launches down the road, OpenAI will only accept agents from people who have verified their identity. Initially, GPT's will be accessible through shareable web links, end quote. The intercept has seen internal documents that suggest crews, new its cars, were struggling to detect large holes, and sometimes had problems recognizing children prior to its California permitting crisis.
Quote, even before its public relations crisis of recent weeks, previously unreported internal materials, such as chat logs show crews has known internally about two pressing safety issues. Driverless crews cars struggled to detect large holes in the road, and have so much trouble recognizing children in certain scenarios that they risked hitting them.
Yet, until it came under fire this month, crews kept its fleet of driverless taxis active, maintaining its regular reassurances of superhuman safety. One of the problems addressed in the internal previously unreported safety assessment materials is the failure of crews' autonomous vehicles to under certain conditions, effectively detect children so that they can exercise extra caution. Quote, crews AVs may not exercise additional care around children, reads one internal safety assessment.
The company's robotic cars, it says, still quote, need the ability to distinguish children from adults so we can display additional caution around children. End quote. In particular, the materials say crews worried its vehicles might drive too fast at crosswalks or near a child who could move abruptly into the street.
The materials also say crews lacks data around kid-centric scenarios, like children suddenly separating from their accompanying adult, falling down, riding bicycles, or wearing costumes. The materials note results from simulated tests in which a crew's vehicle is in the vicinity of a small child.
Quote, based on the simulation results, we can't rule out that a fully autonomous vehicle might have struck the child reads one assessment, and another test drive, a crew's vehicle successfully detected a toddler-sized dummy, but still struck it with its side mirror at 28 miles per hour. The internal materials attribute the robot cars inability to reliably recognize children under certain conditions to inadequate software and testing.
Quote, we have low exposure to small VRUs, vulnerable road users, a reference to children, so very few events to estimate risk from the materials say. Another section concedes crews vehicles, quote, lack of high precision small VRU classifier or machine learning software that would automatically detect child-shaped objects around the car and maneuver accordingly.
The materials say crews, in an attempt to compensate for machine learning shortcomings, was relying on human workers behind the scenes to manually identify children encountered by AVs where its software couldn't do so automatically. Crews has also known that its cars couldn't detect holes, including large construction pits with workers inside, for well over a year according to the safety materials reviewed by the intercept.
Internal crews assessments claim this flaw constituted a major risk to the company's operations. Crews determined that at its current relatively minuscule fleet size, one of its AVs would drive into an unoccupied open pit roughly once a year, and a construction pit with people inside it about every four years. Without fixes to the problems, those rates would presumably increase as more AVs were put on the streets. End quote.
Cold turkey may be great on sandwiches, but there's a better way to break your bad habits. We're talking about our sponsor fume, and they look at the problem a different way. Not everything in a bad habit is wrong. So instead of a drastic uncomfortable change, why not just remove the bad from your habit? Instead of vapor, fume uses flavored air. Instead of electronics, fume is completely natural, and instead of harmful chemicals, fume uses delicious flavors.
You get it, instead of bad, fume is good. It's a habit you're free to enjoy, and makes replacing your bad habit easy. Your fume comes with an adjustable airflow dial and is designed with movable parts and magnets. I told you it's a cool little gadget perfectly balanced, extremely fun to fidget with, which is useful when breaking habits. Stopping is something we all put off because it's hard, but switching to fume is easy, enjoyable, and even fun.
Fume has served over 150,000 customers and has thousands of success stories, and there's no reason that can't be you. Join fume in accelerating humanities, break up from destructive habits, by picking up the journey pack today, head to trifume.com, and use code ride to save 10% off when you get the journey pack today. That's try fum.com and use code ride to save an additional 10% off your order today. If you work in security or IT and your company has AACTA, this message is for you.
Have you noticed that for the past few years, the majority of data breaches and hacks that you read about or hear about on this very podcast have something in common? It's employees, because hackers absolutely love exploiting vulnerable employee devices and credentials. But it doesn't have to be this way. Imagine a world where only one person can access your cloud apps. In this world, fished credentials are useless to hackers, and you can manage every OS, even Linux, from a single dashboard.
Best of all, you can get employees to fix their own device security issues without creating more work for IT. The good news is, you don't have to just imagine this world, you can start living in it just by using collide. Collide is a very important thing to collide. Collide is a device trust solution for companies with AACTA, and it ensures that if a device isn't trusted and secure, it can't log into your cloud apps. Visit collide.com slash ride to watch a demo and see how it works.
That's k-o-l-i-d-e.com slash ride. It's official, we work. Once valued at $47 billion as recently as 2019 has filed for chapter 11 bankruptcy protection for its locations in the US and Canada and has reported liabilities between 10 and 50 billion dollars, which is a lot. Four years ago, within the early years of this very podcast, we work with the highest valued private startup in the world.
Quoting CNBC, we work debuted through a special purpose acquisition company in 2021, but is since lost about 98% of its value. The company in mid-August announced a one-for-forty reverse stock split to get its shares trading back above $1, a requirement for keeping its New York stock exchange listing. We work shares have fallen to a low of about 10 cents, and we're trading at about 83 cents before the stock was halted Monday.
Former co-founder and CEO Adam Newman said that the filing was disappointing. Quote, it has been challenging for me to watch from the sideline since 2019 as we work as failed to take advantage of a product that is more relevant today than ever before. Newman said in a statement to CNBC, I believe that with the right strategy and team, a reorganization will enable we work to emerge successfully. End quote.
From that sad note, I wanted to end today with some signs of green shoots in tech, at least vis-a-vis the idea of a tech IPO pipeline. The sources say, she-in aims for an 80-90 billion dollar valuation in the US IPO that it is planning in the coming months. Far exceeding its 50-60 billion dollar valuation that it's been seeing in private trades, though it is worth remembering that she-in was valued at $100 billion back in 2022, quoting Bloomberg.
While valuation in private trades doesn't necessarily reflect the company's actual valuation, the gap underlines investor concerns over she-in's challenges ranging from intensifying competition to allegations of copyright thefts and potential use of forced labor. It may also complicate she-in's ambitions for a blockbuster listing. She-in was the world's third most valuable startup in 2022 in a funding round value the company at $100 billion.
Its valuation has since dropped along with other startups and technology companies as investors grew wary toward risk assets amid uncertain economic outlooks and higher interest rates. Valuation of bite dance, the parent of short video hit TikTok, fell to below $300 billion in secondary markets in July, down at least 25% from last year.
Bloomberg News has reported, deliberations are ongoing and no final decision has been made regarding she-in's IPO, including its valuation and timing the people said. She-in pioneered ultra-fast fashion, selling new and stylish items such as shirts and swimsuits. For as little as $2 each, it's direct to consumer e-commerce sales took off in the US during COVID and the company quickly became one of the most downloaded shopping apps in the country targeting teens and young women.
The online retailer expects its net income to reach $2.5 billion this year despite the intensifying competition said the people who asked not to be identified as the information is private end quote. In addition to that, by now pay later Fintech, Klarna was also once a supernova expected to have a monster IPO and then, well, the last 18 months happened. But as this piece in TechCrunch points out, Klarna 2 has recently reported a turnaround of sorts.
Numbers that say their business is maybe seeing green shoots as I say. If the high flyers in tech, the ones that were the supernovas that then got arrows in their back, I'm thinking of everyone from Shopify to Roku here. If these companies become recovery stories, then potentially that means we would be seeing signs that the tech recession of the past two years might be ending. Quote, Klarna managed to both increase revenue and swing to profitability in just one year.
Altogether, they're reporting better gross merchandise value, a higher take rate, and solid revenue growth combined with lower cost and smaller credit losses, meaning that Klarna's cost income ratio narrowed to negative 83% in Q3 of 2023 from negative 116% in Q3 of 2022. Boom, profits. The numbers look even better if you allow for adjusted results. Adjusted operating results came in at $43.8 million in the quarter compared to a loss of $143 million a year ago.
It's a good idea to ask just how Klarna pulled this off. Most private companies tend to post huge losses in the name of growth, so it's rare to see a company as large as Klarna, increasing its revenue 30% by 46% year over year. Still, that's only one part of its improving results. Better credit results don't drive revenue. In fact, you could argue that more stringent credit rules could limit growth at a company like Klarna, and the expense of profits, of course.
There's an important factor contributing to the streak of good results. Klarna's bet on the United States is paying off. The company said its US business, Quote, achieved its fourth consecutive quarter of profit with GMV increasing 46% year over year in the third quarter. That's a lot of implied profitable growth and given the relative strength of the US economy compared to other regions, it is likely an important driver of Klarna's current glow-up.
As we are looking forward to an eventual Klarna IPO, we do care about its present value during its last fundraise. Klarna was worth $6.7 billion. So if we annualize its Q3 revenue, we get a figure of around $2.2 billion. Klarna likely has a big holiday quarter underway at the moment, so our revenue figure is likely too small. Still, at our admittedly conservative run rate revenue figure, the company is worth around three exits current top line. Is that a lot? A little?
A look at the trailing price sales multiples of major public fintech companies, Block and PayPal gives us ratios of 1.42x and 2.16x respectively. That said, Block's revenue increased 16% in the third quarter, exclusive a Bitcoin trading revenue, while PayPal's net revenue rose 8% in the last year.
So Klarna is growing its top line faster than either of those two companies and is now paying off the same amount of revenue and either of those two companies and is not priced too far above them if you go by this metric. That allows us to wiggle the math a little and predict that if Klarna can maintain its current trajectory for a quarter or two, it should be able to defend its most recent private market valuation even during a period of depressed fintech prices. And quote.
Nothing for you today, talk to you tomorrow.