by tyler | Apr 5, 2024 | Tech
Faced with labor shortages, sectors such as manufacturing and agriculture are increasingly adopting AI in their automation.
Computer vision startups are looking to jump on that opportunity with a range of point solutions for both industries. From data collection to crop monitoring and harvesting , robots with eyes are entering the fields.
One big challenge that remains, however, is implementation: If such solutions are not easy to use, they won’t be used.
Belgian startup Robovision believes it has found a way around that. The company wants to industrialize deep learning tools and make them more accessible to businesses that are not tech companies at their core. It has built a “no-code” computer vision AI platform that doesn’t require software developers or data scientists to be involved at every step of the process. Robovision doesn’t make robots, but as its name suggests, the company also targets robotics companies that want to develop new machines that support AI-enabled automation.
In practice, this means Robovision customers can use its platform to upload data, label it, test their model and deploy it in production. The company says its model can be useful for a variety of use cases such as recognizing fruit at supermarket scale, identifying faults in newly made electrical components, and even cutting rose stems.
Image Credits: Robovision
Image Credits: Robovision
Out of its base in Belgium, Robovision already serves customers in 45 countries, CEO Thomas Van den Driessche told TechCrunch in an interview. Now, thanks to a recent, sizable funding round, it’s expanding to the U.S., banking on interest from industrial and agribusiness customers in that gigantic market.
The Series A round of $42 million is being co-led by Belgian agtech investor Astanor Ventures and Target Global. The latter is a Berlin-based investor and its participation in this fundraise marks a departure from some of the other coverage it’s had of late: controversy over its ties to Russian money. Red River West, a French VC that focuses on funding European startups looking to break into North America, also participated in the round.
With a post-money valuation of $180 million, this new round brings the total amount of equity funding raised by Robovision to $65 million, including two converted notes. This still leaves the founders together with the staff owning more than 50% of Robovision, its chief growth officer, Florian Hendrickx, told TechCrunch via email.
One challenge that Robovision faces in its expansion is that working with different sectors complicates messaging and its go-to-market strategy. On the plus side, learnings and experiments in one application can be applied to another. Robovision, for example, was able to apply some of the 3D deep learning it had developed for disease detection in tulips to disease detection in human lungs during the Covid crisis.
“It’s a double-edged sword,” founder Jonathan Berte told TechCrunch. “It has been the DNA of Robovision of striking the delicate balance between diversity and focus.”
That DNA comes from Robovision’s history: It was founded in 2012 as a consultancy studio, and it was several years before it pivoted into the B2B platform approach that also made it more attractive to VCs.
The initial traction Robovision gained was in agtech, which represents 50% of its activities, Van den Driessche said. Agtech is also where its Series A’s co-lead investor, Astanor, comes from: That company focuses on what it describes as “impact agrifood.”
Agtech is a sizable opportunity because of labor shortages, and also due to Robovision’s track record — it helps its partner ISO Group plant a billion tulips annually. But other verticals are growing faster for Robovision, Van den Driessche said.
According to Van den Driessche, Robovision is seeing strong traction in life sciences and tech. For instance, Hitachi uses its platform to produce semiconductor wafers. “I don’t think agriculture is going to be the largest sector at scale,” said Bao-Y Van Cong, a partner at Target Global. “I think it’s going to be industrial manufacturing.”
Apple’s recent decision to acquire DarwinAI , an AI startup specializing in overseeing the manufacturing of components, shows rising interest in this space. For Robovision founder Jonathan Berte, it is also a sign that a toolbox that can support a wide variety of different industrialized applications makes more sense. “Apple would never [have bought that] company if it were only a point solution.”
The convertible notes that Robovision raised in 2022 and 2023 following its pivot mostly came from Dutch and Belgian investors, but it had to look further afield to raise the capital it needed. The amount of capital that Robovision raised in the round would have been harder to secure from Benelux, or may have required more dilution.
Robovision’s Belgian roots are paying off in other ways. “The whole early team was very smart people from Ghent university,” Berte said. Van den Driessche became Robovision’s CEO in 2022 , and Berte moved his focus to fundraising, partnerships and global expansion.
Robovision’s tech evolution has extended to rethinking the architecture of its computer vision tools in response to customer demand. Because low latency and delivery speed are requirements in certain environments, it launched Robovision Edge.
In today’s market, doing more with less has become key to competing globally. “I think the only way to do that is to innovate and to become more productive,” Van Cong said.
by tyler | Feb 21, 2024 | Tech
Apple announced today it is upgrading iMessage’s security layer to post-quantum cryptography, starting in iOS and iPadOS 17.4, macOS 14.4, and watchOS 10.4.
The technology giant said that in the coming years, quantum computers will be able to break today’s cryptography standards . That’s why Apple said it is changing how end-to-end encryption works with iMessage without the need for quantum-level processing power.
Today’s messaging apps use encryption typically through a pair of public and private keys. The public key is used to encrypt sent messages and the private key is used by the receiver to decrypt a message, though much of this happens automatically and seamlessly. The cryptography used to scramble user messages today works by applying different math functions. The ability of malicious hackers to decrypt messages relies on the strength of the cryptographic cipher in use today, coupled with the raw computing power aimed at calculating every one of the cipher’s mathematical combinations or permutations.
Apple and other companies believe future quantum computers — capable of exponentially faster computations — could break today’s encryption standards.
“A sufficiently powerful quantum computer could solve these classical mathematical problems in fundamentally different ways, and therefore — in theory — do so fast enough to threaten the security of end-to-end encrypted communications,” Apple said in its blog post .
Apple said that adversaries can start collecting encrypted data today and decrypt it later when quantum computers are more generally available — a technique dubbed “retrospective decryption.”
In its blog, Apple says to protect against future quantum encryption attacks, its encryption keys must change “on an ongoing basis.”
Apple says its new custom built protocol combines Elliptic-Curve cryptography — the existing encryption algorithm for iMessage — and post-quantum cryptography. This forms what Apple calls the PQ3 protocol. When the new PQ3 cryptographic standard rolls out, Apple said it will apply to all new iMessage conversations and older messages by refreshing session keys for prior conversations.
Apple asked two academic research teams to evaluate its PQ3 standard. Since this system is new and we are years away from the general availability of quantum computing power, there is no practical way to measure the efficacy of Apple’s post-quantum protocol.
The tech giant’s announcement comes as a time when lawmakers are looking to introduce online safety rules that run the risk of undermining encryption on messaging services. At the same time, companies like Meta are working on applying end-to-end encryption protection to products like Messenger and Instagram .
End-to-end messaging app Signal last year upgraded to post-quantum encryption algorithms to prevent future quantum-based decryption attacks.
Apple fixes bug that undermined iOS privacy feature for years
by tyler | Jan 24, 2024 | Tech
European Union lawmakers scrambling for the bloc to be a contender in the generative AI race are presenting a package of support measures aimed at charging up homegrown AI startups and scale ups.
Artificial intelligence technologies — and especially generative AI models which are trained on very large data-sets and have capabilities such as being able to parse natural language and produce text, imagery or audio on demand — are being viewed as a key strategic area for the bloc’s future competitiveness. However Commission officials concede lawmakers have been caught on the hop, somewhat, when it comes to compute infrastructure that’s fit for training such AIs.
They admit they were taken by surprise by the sudden rise of generative AI tools like OpenAI’s ChatGPT last year. So while the EU boasts an impressive network of high performance supercomputers — such as the newly inaugurated MareNostrum 5 — this strategic infrastructure, historically been geared towards scientific users (and use-cases), hasn’t been optimized for training the new generation of disruptive generative AI models. And that’s one big missing component in the EU’s AI strategy that lawmakers are now rushing to change.
The “AI innovation package”, as the policy bundle that’s been adopted by the Commission today is being dubbed, seeks to extend support for growing Europe’s AI ecosystem with interventions across a number of areas — including high performance computing infrastructure for training models; and access to the necessary skills and talent to make the generative AI magic happen.
Commenting in a statement, Margrethe Vestager, the EU’s digital chief, said: “You need computing power to develop AI. A lot of it. So we want to give SMEs and start-ups privileged access to the network of European supercomputers. We are committed to innovation of AI and innovation with AI. And we will do our best to build a thriving AI ecosystem in Europe.”
“Today, we announce the launch of AI Factories, bringing together the ‘raw materials’ for AI: computing power, data, algorithms and talent,” added internal market commissioner, Thierry Breton, in another supporting statement. “They will serve as a one-stop shop for Europe’s AI start-ups, enabling them to develop the most advanced AI models and industrial applications. We are making Europe the best place in the world for trustworthy AI.”
Back in November, the Commission already launched the Large AI Grand Challenge , a prize to give AI startups financial support and supercomputing access. “Today’s package puts this commitment into practice through a broad range of measures to support AI startups and innovation, including a proposal to provide privileged access to supercomputers to AI startups and the broader innovation community,” the Commission suggested.
The EU’s executive is trailing the idea of setting up so-called “AI Factories” around the bloc’s supercomputers — via a blended bundle of policy interventions that will usher in AI-focused hardware upgrades; measures to boost access to the resource, including for startups; a support “one-stop-shop” for startups and innovators to access “supercomputer-friendly programming facilities and other AI enabling services”; and additional support measures aimed at fostering the development of AI applications based on General Purpose AI models.
“We have hundreds of promising startups in generative AI and in related application areas but they need access to this computing power and they also need access to other key ingredients of AI,” said a Commission official, briefing the press on background ahead of today’s announcements. “It starts with data, then it’s the computing power to train the models based on that data and you need the algorithm of course but the fourth ingredient, I would say, is talent and skills.”
On compute infrastructure, the EU’s executive has proposed amending existing EU legislation related to the joint procurement of high performance supercomputers (aka, the Regulation establishing the European High Performance Computing Joint Undertaking ) to increase flexibility around upgrades.
Commission officials say the proposed amendment to the rules is intended to enable the bloc to ramp up the available capacity of its supercomputers far more flexibly than is possible under the current framework.
“It is much more cost efficient to double the capacity of the existing supercomputers, by adding AI capabilities now, rather than wait for another two or three years. Til we have completed the procurement of a new AI dedicated supercomputer,” noted a second Commission official, also briefing the press on background.
The support package is intended to complement the bloc’s nascent regulatory framework for AI — aka, the AI Act — which EU lawmakers also present as a supportive measure. But where that law aims to boost adoption of AI apps by fostering user trust, today’s package is all about more direct forms of support for AI app makers.
The AI Act has also faced some industry push-back. Last year saw especially fierce lobbying for regulatory carve-outs for general purpose AI from the likes of France’s Mistral . So the Commission may be feeling under industry pressure to pull a few juicy-looking carrots out of their hat too. (Asked for his take on the EU’s AI support package, Mistral CEO and co-founder, Arthur Mensch, gave the strategy a thumbs up, saying: “It works well.”)
Add to that, while a political deal on the AI Act was clinched last month the law remains in draft form, pending a vote in the Council to affirm the compromise text settled on then — so there’s still a chance of a spanner being thrown in the works of the flagship AI legislation, even at this late stage.
In public, the Commission is projecting confidence the AI Act will sail through these final hoops and get adopted. In private, Commission officials offer the slightly more cautious message that they are “hopeful” the regulation will be adopted.
Nonetheless, today the EU’s executive is publicly pressing ahead in the expectation of the flagship AI law setting sail by taking steps to establish a new public body, called the AI Office, which will be tasked with enforcing the Act on general purpose AIs.
The AI Office, which will be set up within the Commission, will have a number of other roles too. The Commission envisages it as a “central coordination body” for AI policy at EU level — cooperating with other departments, EU bodies, Member States and the stakeholder community.
“It will have an international vocation and promote the EU approach to AI governance and contribute to the EU’s international activities on AI,” it writes in a press release . “More generally, the AI Office should build up knowledge and understanding on AI and foster AI uptake and innovation.”
The decision to establish the AI Office enters into force today — with operations expected to commence in the following months. But the proposed tweaks to supercomputing procurement rules will need the backing of the European Parliament and Council to proceed.
While having a network of existing supercomputers for generative AI model training might sound like a handy strategic power-up in the global AI race, the bloc’s supercomputers are already overbooked by a factor of at least two, per Commission officials. They were also not designed with this kind of AI model training in mind. So upgrades — both hardware and capacity based — are sorely needed if the resource is to level up the bloc’s prospects here.
As we’ve reported before , the EU has also recognized that AI startups will need a fair amount of hand-holding to get the most out of the resource. Hence support services being another focus of today’s package.
Research uses of the bloc’s supercomputing infrastructure are free and AI startups doing model pre-training or training will also be able to benefit from free access. So the planned investments to beef up the EU’s supercomputers and configure the infrastructure for training generative AIs could, in theory, help level the competitive playing field — instead of the lumpy reality we have now, where startups with a cosy partnership with a hyperscaler and their cloud computing infrastructure are able to run ahead of the pack.
At the same time, scores of AI startups wanting to tap into free EU compute infrastructure will clearly amp up demand for slots on the supercomputers. So the success of the EU’s supercomputing-for-AI strategy will rest on delivering key infrastructure upgrades — and fast.
The Commission’s plan to ease the demand bottleneck is to amend current rules to allow greater freedom to upgrade the existing supercomputers. These upgrades will be focused on adapting the infrastructure so it’s better suited to the needs of generative AI model developers. So upgrades in this context basically means adding lots and lots of racks packed with accelerators (i.e. GPUs). Plus some rewiring of components to ensure data can get where it needs to go efficiently.
“With this flexibility on upgrading the existing infrastructure, within a year, we can almost double the capacity and provide the AI users the access that they need to train large models and stay competitive on a very fast evolving market in the world,” one Commission official suggested.
Of course the EU can tweak legislation to enable more flexible upgrades to its supercomputers but it can’t guarantee it will be able to acquire all the necessary GPUs — given it will be competing with extremely high global demand for the components.
Asked about this during the press briefing, Commission officials admitted it’s a big worry — and said delays are possible, given so much demand for the chips. But they also noted the bloc has been seeking commitments from leading GPU maker Nvidia in relation to supercomputer procurement.
Over the longer term, the EU is also hoping other measures it’s taking to bolster access to chips, such as the Chips Act , will enable a European supply of accelerators to ease the demand crunch. But the official conceded its near term ambitions to make its high performance compute infrastructure fit for the generative AI boom may falter over chip supply issues. “There’s no guarantee we will not suffer similar shortage or delay as other clients for Nvidia. But we’re working on a number of mitigation measures, short term or long term,” they said.
Elsewhere, the “AI innovation” package aims to smooth the road for European generative AI startups to scale in a number of ways.
For example, EU Member States wanting to host a dedicated AI-optimized supercomputer will need to commit to providing a direct connection to data centers to ensure “quick and easy access” to the infrastructure, per the Commission. They will also need to commit to providing a wrapper of support services to ensure startups get the necessary help to make use of the infrastructure. “All this should be accessible, as a one stop shop, for any user wherever he is residing in the Union,” the official added.
There is also going to be some extra financial support made available for AI startups — via the existing Horizon Europe and the Digital Europe programs — specifically focused on generative AI. The Commission said it expects this to generate an additional overall public and private investment of around €4BN until 2027.
Plus the EU says it will seek to drive further public and private investments in AI start-ups and scale-ups, including through venture capital or equity support (such as via new initiatives of the EIC accelerator Programme and InvestEU ).
Tackling the data piece of the AI puzzle, the support package is geared towards accelerating the development and deployment of the Common European Data Spaces , which were set out in the Commission’s February 2020 data strategy , to made them available to the “AI community”.
The Commission is also flagging a “GenAI4EU” initiative to support the development of “novel use cases and emerging” AI applications in Europe’s 14 industrial ecosystems, as well as the public sector. “Application areas include robotics, health, biotech, manufacturing, mobility, climate and virtual worlds,” it said.
In a further step, two so-called European Digital Infrastructure Consortiums (or Edics) are being established — one of which, called the Alliance for Language Technologies, will be focused on developing a common European infrastructure in language technologies with the aim of tackling a shortage in European languages data for training AIs. The goal is to support the development of European large language models (LLMs), and uphold Europe’s “linguistic diversity and cultural richness”, as the Commission puts it.
The second Edic, called CitiVerse, will aim to support development of AI tools focused on city planning and management — aka “Local Digital Twins for Smart Communities”, as the PR has it — with the goal of leveraging generative AI to help cities model and optimise processes such as traffic management and waste management, or see how a proposed development might impact an urban area.
The bloc’s president, Ursula von der Leyen, trailed the plan to let AI startups tap the EU’s supercomputers for model training last fall . At the time she said the access would be provided to “responsible” AI startups — which agree to abide by the EU’s risk-based governance model for AI applications.
The Commission’s PR today conveys a similar claim — with the announcement that the package of measures will support European startups and SMEs “in the development of trustworthy Artificial Intelligence (AI) that respects EU values and rules”. And the word “trustworthy” occurs several more times in the announcement.
Thing is, the EU hasn’t yet passed the AI Act — and, even if it’s adopted soon, the law won’t be fully in force for several years — so how will AI startup trustworthiness be determined, in order to decide who gets supercomputer access, in the meanwhile?
The bloc has been working with industry on a stand-in (non-legally binding) AI Pact — which is intended to act as a stop-gap code until the AI Act is in force. And one of the ways AI startups that want to be able to tap free training time on the supercomputers will be able to signal they’re appropriately “trustworthy” is by signing up to this Pact, per the Commission.
However, as we reported last month , the EU is already providing some AI startups with slots on the supercomputers — such as France’s Mistral — even though the AI Pact isn’t technically up and running yet… So there appears to be a bit of a gap between the bloc’s publicity, about access for “responsible/trustworthy” AI startups, and how the early phases of this program are operating.
Pressure for the EU to play catch-up in generative AI appears to be the main driving force behind its early moves here, while hard AI governance rules are still generating (ha!) debate. But if the AI Act fails to make it over the line, the AI Pact will be no better than a ‘pinkie promise’, and the EU’s own pledge to funnel taxpayer-funded support only at “trustworthy” AIs will ring hollow.
“For the AI Pact — really — to become active we first needed to have the full text of the AI Act,” a Commission official told TechCrunch when asked about this. “Now that we have made this big step of the political agreement [on the Act] reached in December — and hoping for adoption soon — this is really a moment where we would start operationalizing the AI Pact further — seeking further commitments from companies and from startups.”
“When these AI dedicated supercomputers become operational and these AI factories become operational then, indeed, the AI Pact can be used as one of the ways to to demonstrate that the proposal is in line with EU rules and EU values,” they added.
EU to let ‘responsible’ AI startups train models on its supercomputers
EU to expand support for AI startups to tap its supercomputers for model training
by tyler | Jan 17, 2024 | Tech
Anyone who knows your WhatsApp number can figure out if you are only using the mobile app, or its companion web or desktop apps, a security researcher found.
Tal Be’ery, the co-founder and CTO of crypto wallet maker ZenGo , found that it’s possible to determine whether a user on WhatsApp is using more than just the mobile app. Be’ery demonstrated and proved his findings in tests performed with WhatsApp numbers controlled by TechCrunch.
While revealing where users have WhatsApp running is not the most dangerous leak of information, digital security experts agree that it’s not an ideal situation, and, in some cases, it could help hackers target WhatsApp users.
“[It] could be useful for information gathering and plotting an attack,” Runa Sandvik, a digital security expert, told TechCrunch, referring to how hackers could figure out that their target is using WhatsApp on a desktop, which is generally an easier target to compromise than a mobile phone.
“It at least tells you more about the devices they use and how ‘accessible’ their WhatsApp setup may be,” said Sandivk, who is the founder of Granitt , a startup that aims to train at-risk people like journalists, activists, and politicians.
Meta’s spokesperson Zade Alsawah told TechCrunch that the company received Be’ery’s research and concluded that the app’s current design “is what users want and expect.”
“It used to be the case that your phone had to be online to receive messages and that provided significant limitations for people. With multi device users can send and receive their personal messages across devices privately with end-to-end encryption — and that’s the direction we’ll continue to take,” Alsawah said in a statement.
Harlo Holmes, the chief information security officer and director of digital security at the Freedom of the Press Foundation, said that being able to tell what devices people are using WhatsApp on is a privacy issue.
Referring to the ability to disable read receipts and typing indicators on WhatsApp, Holmes said that WhatsApp should offer a similar opt-out feature for device indicators.
“Presence-related metadata should be protected and opt-in. Similar to geolocation, away status, and read receipts; this is no different,” Holmes told TechCrunch.
In practice, Holmes said, “perhaps a stalker could deduce that I’m at home or not, depending on which device I used.”
Be’ery wrote in his blog post explaining the data leak that it is a consequence of the way WhatsApp is designed: When someone sends a message to another WhatsApp user, their device creates a different session key for each device the receiver is using, thus telling the sender how many devices the receiver is using.
Anyone can find out this kind of information by using WhatsApp on the web and inspecting traffic with a browser’s developer tool, Be’ery explained. The only thing a malicious attacker has to do to find out this information is to add the target to their contact list, and this works even if the target blocks the attacker’s number, as Be’ery demonstrated to TechCrunch.
In other words, there is nothing a person can do to prevent others from seeing this type of information. And WhatsApp isn’t going to change how the app works either — at least for now.
PSA: Your chat and call apps may leak your IP address
