Connect with us


OpenAI’s DALL-E creates plausible images of literally anything you ask it to



OpenAI’s latest strange yet fascinating creation is DALL-E, which by way of hasty summary might be called “GPT-3 for images.” It creates illustrations, photos, renders, or whatever method you prefer, of anything you can intelligibly describe, from “a cat wearing a bow tie” to “a daikon radish in a tutu walking a dog.” But don’t write stock photography and illustration’s obituaries just yet.

As usual, OpenAI’s description of its invention is quite readable and not overly technical. But it bears a bit of contextualizing.

What researchers created with GPT-3 was an AI that, given a prompt, would attempt to generate a plausible version of what it describes. So if you say “a story about a child who finds a witch in the woods,” it will try to write one — and if you hit the button again, it will write it again, differently. And again, and again, and again.

Some of these attempts will be better than others; indeed, some will be barely coherent while others may be nearly indistinguishable from something written by a human. But it doesn’t output garbage or serious grammatical errors, which makes it suitable for a variety of tasks, as startups and researchers are exploring right now.

DALL-E (a combination of Dali and WALL-E) takes this concept one further. Turning text into images has been done for years by AI agents, with varying but steadily increasing success. In this case the agent uses the language understanding and context provided by GPT-3 and its underlying structure to create a plausible image that matches a prompt.

As OpenAI puts it:

GPT-3 showed that language can be used to instruct a large neural network to perform a variety of text generation tasks. Image GPT showed that the same type of neural network can also be used to generate images with high fidelity. We extend these findings to show that manipulating visual concepts through language is now within reach.

What they mean is that an image generator of this type can be manipulated naturally, simply by telling it what to do. Sure, you could dig into its guts and find the token that represents color, and decode its pathways so you can activate and change them, the way you might stimulate the neurons of a real brain. But you wouldn’t do that when asking your staff illustrator to make something blue rather than green. You just say, “a blue car” instead of “a green car” and they get it.

So it is with DALL-E, which understands these prompts and rarely fails in any serious way, although it must be said that even when looking at the best of a hundred or a thousand attempts, many images it generates are more than a little… off. Of which later.

In the OpenAI post, the researchers give copious interactive examples of how the system can be told to do minor variations of the same idea, and the result is plausible and often quite good. The truth is these systems can be very fragile, as they admit DALL-E is in some ways, and saying “a green leather purse shaped like a pentagon” may produce what’s expected but “a blue suede purse shaped like a pentagon” might produce nightmare fuel. Why? It’s hard to say, given the black-box nature of these systems.

But DALL-E is remarkably robust to such changes, and reliably produces pretty much whatever you ask for. A torus of guacamole, a sphere of zebra; a large blue block sitting on a small red block; a front view of a happy capybara, an isometric view of a sad capybara; and so on and so forth. You can play with all the examples at the post.

It also exhibited some unintended but useful behaviors, using intuitive logic to understand requests like asking it to make multiple sketches of the same (non-existent) cat, with the original on top and the sketch on the bottom. No special coding here: “We did not anticipate that this capability would emerge, and made no modifications to the neural network or training procedure to encourage it.” This is fine.

Interestingly, another new system from OpenAI, CLIP, was used in conjunction with DALL-E to understand and rank the images in question, though it’s a little more technical and harder to understand. You can read about CLIP here.

The implications of this capability are many and various, so much so that I won’t attempt to go into them here. Even OpenAI punts:

In the future, we plan to analyze how models like DALL·E relate to societal issues like economic impact on certain work processes and professions, the potential for bias in the model outputs, and the longer term ethical challenges implied by this technology.

Right now, like GPT-3, this technology is amazing and yet difficult to make clear predictions regarding.

Notably, very little of what it produces seems truly “final” — that is to say, I couldn’t tell it to make a lead image for anything I’ve written lately and expect it to put out something I could use without modification. Even a brief inspection reveals all kinds of AI weirdness (Janelle Shane’s specialty), and while these rough edges will certainly be buffed off in time, it’s far from safe, the way GPT-3 text can’t just be sent out unedited in place of human writing.

It helps to generate many and pick the top few, as the following collection shows:

AI-generated illustrations of radishes walking dogs.

The top 8 out of a total of X generated, with X increasing to the right.

That’s not to detract from OpenAI’s accomplishment here. This is fabulously interesting and powerful work, and like the company’s other projects it will no doubt develop into something even more fabulous and interesting before long.

Lyron Foster is a Hawaii based African American Musician, Author, Actor, Blogger, Filmmaker, Philanthropist and Multinational Serial Tech Entrepreneur.


Google claims almost no change in ad revenue from targeting proposals in its Privacy Sandbox — but privacy upside less clear



As Google’s Privacy Sandbox remains under scrutiny over competition concerns, the tech giant has released an update claiming experimental ad-targeting techniques it’s developing as part of the plan to depreciate support for third party cookies on its Chrome browser show results that are “nearly as effective as cookie-based approaches”.

Google has been working on a technique — called Federated Learning of Cohorts (FLoC) — to target ads based on clustering users into groups with similar interests, which it claims is superior from a privacy perspective vs the current (dysfunctional) ‘norm’ of targeting individuals based on third parties tracking everything they do online.

It wants FLoCs to enable interest-based advertising to continue after it ends support for third party trackers.

However the proposal has alarmed advertisers who argue it’s anti-competitive. And earlier this month the UK’s Competition and Markets Authority (CMA) opened an investigation of the Privacy Sandbox proposal after complaints from a coalition of digital marketing companies and others from newspapers and technology companies alleging Google is abusing a dominant position by depreciating support for third party trackers.

On the privacy front Google’s self-styled Privacy Sandbox isn’t exactly attracting effusive plaudits, either.

The Electronic Frontier Foundation has, for example, dubbed FLoCs “the opposite of privacy-preserving technology” — warning in 2019 that the approach is akin to a “behavioral credit score”. It said then that the proposals risk sustaining discrimination against vulnerable groups of people, whose online activity would be pattern-matched with others without their say-so; and could also lead to leaking sensitive info about them to third parties — without offering web users any way to escape being put in a ‘interest based’ ad targeting bucket. 

With objections piling up from on sides of the aisle (advertiser vs user) — and now active regulatory scrutiny of the competition issue — Google has its work cut out to sell its preferred replacement for tracking cookies to all the relevant stakeholders. Though advertisers (and competition regulators) currently seem front of mind for the tech giant.

In an update about the Privacy Sandbox proposals today, Google appears to be hoping to alleviate advertisers’ concerns that the demise of tracking cookies will degrade their ability to lucratively target Internet users — writing that tests of the FLoC technology suggest advertisers will continue to see “at least 95% of the conversions per dollar spent when compared to cookie-based advertising”.

It’s not clear how much test data was involved in Google generating that percentage, however. (We asked and Google did not have an immediate response.) So there’s zero meat on the bone of the ‘95% minimum’ claim.

Its spokesman did note that it will be opening up public testing in March — and expects advertisers to join in kicking FLoC’s tires then. So there’s clearly going to be more detail to come on this front.

“Chrome intends to make FLoC-based cohorts available for public testing through origin trials with its next release in March and we expect to begin testing FLoC-based cohorts with advertisers in Google Ads in Q2,” writes Chetna Bindra, group product manager for user trust and privacy in the blog post, adding: “If you’d like to get a head start, you can run your own simulations (as we did) based on the principles outlined in this FLoC whitepaper.”

It’s unsurprisingly that Google continues to emphasize the relative openness with which it’s developing the Privacy Sandbox proposals — as that may help it fight antitrust accusations. But it’s also noteworthy being as the adtech industry, which has been fighting to block/delay its depreciation of third party cookies, is busy spinning up its own contenders to replace trackers — and developing those competing proposals typically with a lot less transparency than Google.

Nonetheless, Google seems a whole more comfortable quantifying FLoC’s potential impact on ad revenue (tiny, per its latest claim) vs articulating what privacy gains Internet users might expect from the proposed shift from individual tracking to run behavioral ads to being stuck in labelled buckets to run behavioral ads.

Google’s blog post has a few fuzzy mentions — like “viable privacy-first alternatives” and ‘hiding individuals “in the crowd”’ — but there’s no metric or data offered on how much privacy users stand to gain if its preferred post-cookie future comes to pass.

Test results it published in October also focused on seeking to demonstrate to advertisers that FLoCs can deliver on other relevant ad metrics. Funnily enough, Internet users’ privacy — and what happens when degrees of privacy are lost — is rather harder for Google’s computer scientists to measure.

“The idea is to make it so that no one can reconstruct your cross-site browsing history,” said the company’s spokesman when we asked about how the proposal will improve users’ privacy standing.

“We’re trying to address non-transparent forms of tracking, across websites, with privacy-safe mechanisms for consumers, and make it so it can’t happen. And to do so while still enabling opportunity and fair compensation for publishers and advertisers. So it’s really not even a matter of trying to approximate a kind of privacy: We’re trying to address a root critical concern of users, full stop,” he added.

FLoCs are just one part of Google’s Privacy Sandbox proposals. The company is working on a slew of aligned efforts to simultaneously replace various other key components of the adtech ecosystem. And it gives an overview of some of these in the blog post — covering proposals for (post-cookie) conversation measurement; ad-fraud prevention; and anti-fingerprinting.

Here it dwells briefly on retargeting/remarketing — referring to a new Chrome proposal (called Fledge) that it says it’s considering for a ‘trusted server’ model “specifically designed to store information about a campaign’s bids and budgets”. This will also be made available for advertisers to test later this year, Google adds.

“Over the last year, several members of the ad tech community have offered input for how this might work, including proposals from Criteo, NextRoll, Magnite and RTB House. Chrome has published a new proposal called FLEDGE that expands on a previous Chrome proposal (called TURTLEDOVE) and takes into account the industry feedback they’ve heard, including the idea of using a ‘trusted server’ — as defined by compliance with certain principles and policies — that’s specifically designed to store information about a campaign’s bids and budgets. Chrome intends to make FLEDGE available for testing through origin trials later this year with the opportunity for ad tech companies to try using the API under a “bring your own server” model,” it writes.

“Technology advancements such as FLoC, along with similar promising efforts in areas like measurement, fraud protection and anti-fingerprinting, are the future of web advertising — and the Privacy Sandbox will power our web products in a post-third-party cookie world,” it adds.

Discussing Fledge’s potential, Dr Lukasz Olejnik, an independent researcher and consultant, said there’s still a lot of uncertainty over how it might impact user privacy. “The Fledge experiment looks potentially interesting but it mixes in various proposals in this test. Such a mix would need to get a specific privacy assessment as the offered privacy qualities might be different than original claimed. Furthermore, the current tests will have many privacy precautions intended for the future, turned off initially. It will be tricky to gradually turn them on,” he told TechCrunch.

Continue Reading


Unpacking Chamath Palihapitiya’s SPAC deals for Latch and Sunlight Financial



This morning, investor and SPAC raconteur Chamath Palihapitiya announced two new blank-check deals involving Latch and Sunlight Financial.

Latch, an enterprise SaaS company that makes keyless entry systems, has raised $152 million in private capital, according to Crunchbase. Sunlight Financial, which offers point of sale financing for residential solar systems, has raised north of $700 million in venture capital, private equity and debt.

We’re going to chat about the two transactions.

There’s no escaping SPACs for a bit, so if you are tired of watching blind pools rip private companies into the public markets, you are not going to have a very good next few months. Why? There are nearly 300 SPACs in the market today looking for deals, and many will find one.

The Exchange explores startups, markets and money. Read it every morning on Extra Crunch, or get The Exchange newsletter every Saturday.

Think of SPACs are increasingly hungry sharks. As a shark get hungrier while the clock winds down on its deal-making window, it may get less choosy about what it eats (take public). There are enough SPACs on the hunt today that they would be noisy even if they were not time-constrained investment vehicles. But as their timers tick, expect their dealmaking to get all the more creative.

This brings us back to Chamath’s two deals. Are they more like the Bakkt SPAC, which led us to raise a few questions? Or more akin to the Talkspace SPAC, which we found pretty reasonable? Let’s find out.

Keyless locks = Peloton for real estate

Let’s start with the Latch deal.

New York-based Latch sells “LatchOS,” a hardware and software system that works in buildings where access and amenities matter. Latch’s hardware works with doors, sensors and internet connectivity.

The company has raised a number of private rounds, including a $126 million deal in August of 2019 which valued the company at $454.3 million on a post-money basis, according to PitchBook data. The company raised another $30 million in October of 2020, though its final private valuation is not known.

As Chamath tweeted this morning, Latch is merging with TS Innovation Acquisitions Corp, or $TSIA. The SPAC is associated with Tishman Speyer, a commercial real estate investor. You can see the synergies, as Latch’s products fit into the commercial real estate space.

Up front, Latch is not a company that is only reporting future revenues. It has a history as an operating entity. Indeed, here’s its financial data per its investor presentation:

Doing some quick match, Latch grew 50.5% from 2019 to 2020. Its software revenues grew 37.1%, while its hardware top line expanded over 70% during the same period. So, the company’s revenue mix shifted more towards hardware incomes in 2020.

That could be due to strong hardware installation fees, which could later result in software revenues; the company claims an average of a six-year software deal, so hardware revenues that are attached to new software incomes could lowkey declaim long-term SaaS revenues.

While some were quick to note that the company is far from pure-SaaS — correct — I suspect that the model that will get some traction amongst investors is that this feels a bit like Peloton for real estate. How so? Peloton has large hardware incomes up-front from new users, which convert to long-term subscription revenues. Latch may prove similar, albeit for a different customer base and market.

Per the deal’s reported terms, Latch will be worth $1.56 billion after the transaction. And the combined entity will have $510 million in cash, including $190 million from a PIPE — a method of putting private money into a public entity — from “BlackRock, D1 Capital Partners, Durable Capital Partners LP, Fidelity Management & Research Company LLC, Chamath Palihapitiya, The Spruce House Partnership, Wellington Management, ArrowMark Partners, Avenir and Lux Capital.”

Continue Reading


Calling Swedish VCs: Be featured in The Great TechCrunch Survey of European VC



TechCrunch is embarking on a major project to survey the venture capital investors of Europe, and their cities.

Our survey of VCs in Stockholm, and Sweden generally, will capture how the country is faring, and what changes are being wrought amongst investors by the coronavirus pandemic.

The deadline is the end of this week.

We’d like to know how Sweden’s startup scene is evolving, how the tech sector is being impacted by COVID-19, and, generally, how your thinking will evolve from here.

Our survey will only be about investors, and only the contributions of VC investors will be included. More than one partner is welcome to fill out the survey. (Please note, if you have filled the survey out already, there is no need to do it again).

The shortlist of questions will require only brief responses, but the more you can add, the better.

You can fill out the survey here.

Obviously, investors who contribute will be featured in the final surveys, with links to their companies and profiles.

What kinds of things do we want to know? Questions include: Which trends are you most excited by? What startup do you wish someone would create? Where are the overlooked opportunities? What are you looking for in your next investment, in general? How is your local ecosystem going? And how has COVID-19 impacted your investment strategy?

This survey is part of a broader series of surveys we’re doing to help founders find the right investors.

For example, here is the recent survey of London.

You are not in Sweden, but would like to take part? That’s fine! Any European VC investor can STILL fill out the survey, as we probably will be putting a call out to your country next anyway! And we will use the data for future surveys on vertical topics.

The survey is covering almost every country on in the Union for the Mediterranean, so just look for your country and city on the survey and please participate (if you’re a venture capital investor).

Thank you for participating. If you have questions you can email

(Please note: Filling out the survey is not a guarantee of inclusion in the final published piece).

Continue Reading