Every time you use ChatGPT or generate an image with AI, there is a memory chip working at extreme speed behind the scenes. However, that chip has a memory bottleneck problem, and a Korean research team may have just solved it.

Researchers at POSTECH (Pohang University of Science and Technology) developed a new way to stack more than 10 ultrathin semiconductor chips on top of each other, achieving a memory density roughly four times higher than the best commercial chips available today (via TechXplore).

Why is stacking chips so hard, and what makes this one different?

High-bandwidth memory, or HBM, is the type of memory that powers AI accelerators. It works by stacking multiple chips vertically, much like building a high-rise instead of spreading out across land.

The problem is that as chips get thinner, they become incredibly fragile. At one-fifth the thickness of a human hair, they bend, warp, and crack under pressure. Current manufacturing methods make this worse, often damaging chips before they even make it into a stack.

The POSTECH team solved this by combining two techniques into one process. Transfer printing precisely places each chip where it needs to go, while in-situ bonding forms the metallic connections at the same moment, all under low heat below 180 degrees Celsius and low pressure below 20 kilopascals. The result is a stack of more than 10 chips with almost no misalignment and very little warping.

Why this matters for the future of AI

More memory packed into the same space means AI tools can run faster and handle bigger tasks without needing larger or more expensive hardware. The researchers also see uses beyond AI, including next-generation micro-LED displays and advanced processor designs that need the same kind of ultra-precise stacking this method delivers.

Getting this into commercial production is the next step, but if it gets there, the memory ceiling that has been quietly holding AI back could finally start to lift.

Uninstalling apps on macOS is usually very easy. You drag an app to the Trash, empty it, and move on. The annoying part is that many apps still leave residue behind, including support files, caches, preferences, containers, and logs. I have always found that frustrating, especially when old app data keeps sitting around long after the app itself is gone.

AppCleaner by FreeMacSoft has been the popular go-to option for this for years, and it still does the job well. But I recently came across a new open-source alternative called Uninstally by Codenta, which solves the same basic problem. It removes Mac apps along with the support files, caches, preferences, containers, logs, and other leftovers they usually leave behind.

How does Uninstally work?

Uninstally can be used directly from Finder. Once its Finder extension is enabled, you can right-click any .app bundle and choose “Uninstall with Uninstally.” The app then opens a confirmation window instead of making you start from a separate app browser.

The cleaner part is how it finds related files. Uninstally uses the app’s bundle identifier and helper namespaces to match leftover items across the Library hierarchy, rather than just looking for folders with the same name. Before anything is removed, it shows the app name, icon, reclaimable storage, item count, and lets you review or deselect matched files.

What else makes it useful?

There is also a standalone app browser for a more deliberate cleanup. You can search installed apps, switch between grid and list views, and filter by largest apps, recently installed apps, never opened apps, broken installs, duplicated apps, and apps with leftovers.

Uninstally also includes a leftover scanner for apps you have already removed. Instead of digging through Library folders manually, you can scan for orphaned support files, caches, containers, preferences, logs, and old installers in one place.

It also supports Homebrew casks and formulae, shows dependency relationships, and can remove Homebrew leftovers through optional zap cleanup. User-domain files are moved to the Trash, while privileged items require an administrator prompt. You can download Uninstally from Codenta’s website or its GitHub repo.

Google is rolling out a feature that flags when an advertisement was made using AI. The label will indicate if an ad was created or edited with generative tools, TechCrunch reports.

The disclosure appears in the “My Ad Center” panel, reachable via the three-dot menu or info icon on ads. It covers ads across Google Search, YouTube, and Google Discover, and is available globally.

That panel already lets users block or report ads and learn why one was shown. Now it adds an option labelled “how this ad was made”, which surfaces any AI involvement.

The rationale is straightforward. AI makes it cheap to generate slick product imagery, which can mislead shoppers who assume they are looking at a real photograph rather than a synthetic one.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Until now, Google only required AI disclosure on election ads. Extending it to commercial ads is a meaningful widening of the policy.

The honour-system catch

The reach of the feature depends heavily on how an ad was built. When advertisers use Google’s own generative AI ad tools, the disclosure is switched on automatically.

When an ad is made elsewhere, though, the advertiser must actively flag that AI was involved. Google says it will not run its own check to verify the claim, so the label rests on advertisers being honest.

That gap matters because the incentive to stay quiet is real. An advertiser hoping a synthetic scene passes for a genuine photo has little reason to volunteer otherwise, and Google is not looking over its shoulder.

Regulators are forcing the issue

The timing is not accidental. Google’s move front-runs tougher rules, as the EU AI Act’s transparency obligations for AI-generated content start to bite in August.

Industry is already resisting the mandatory version, with retailers lobbying to exempt AI-made ads from those EU rules. A voluntary, self-declared label is a far lighter touch than what Brussels has in mind, and part of a broader fight over the AI Act.

Google is not consistent across its own products either. On YouTube it will auto-label AI videos whether or not creators disclose them, a stricter stance than the advertiser honesty it relies on here.

Transparency, up to a point

The feature is still a step toward a market drowning in synthetic media, where even Google has branded some AI content spam. Giving users a place to ask how an ad was made is better than silence.

Whether it changes behaviour is another question, in an ecosystem where deceptive advertising is already a lucrative problem. A label only helps if the people with the most to hide choose to apply it.

For now, Google has built the disclosure and handed advertisers the switch. The honest ones will flip it, and the rest are exactly the reason such a label was needed.

Ireland, Spain, France and the Netherlands are the only member states yet to incorporate the NIS2 directive into national law.

Ireland is one of four countries being referred to the highest court in the European Union for failing to adopt cybersecurity directives into law. The European Commission’s move comes as Ireland commences its six-month rotational presidency heading the EU Council.

The Network and Information Security 2 (NIS2) Directive entered into force in January 2023 and sets high security standards across 18 critical sectors, including health, energy, transport and the public sector, mandating organisations to implement appropriate security measures and report any relevant incidents to the authorities.

However, directives must be incorporated into national legislation by EU member states before gaining effect. Member states had until October 2024 to carry out the transposition.

But by late November 2024, 23 member states, including Denmark, Germany, Finland and Sweden, were yet to transpose the directive, while by May 2025, 19 had still not done so.

In its referral yesterday (8 July), which also includes Spain, France and the Netherlands, the Commission requested the Court of Justice of the European Union to impose financial sanctions on infringing member states, consisting of lump sum and daily penalties until NIS2 is incorporated into national legislation.

The cybersecurity threat landscape is fast evolving, as newer technologies such as AI provide bad actors with advanced tools to commit phishing attacks, scams and infrastructure break-ins, while breaches go underreported in Ireland, according to a recent Compliance Institute report.

“While Ireland is not alone in having missed the deadline, this is not a great start for Ireland to our presidency of the Council of the European Union,” said Dentons partner David Kirton.

“The Government has listed competitiveness and security as two of its three key pillars for the presidency, so putting this legislation into effect would be a strong symbol of that commitment.”

The Government published a general scheme of the National Cyber Security Bill in August 2024, and a National Digital and AI Strategy this February, where it committed to “prioritising legislation to implement the EU NIS2 Directive”, but did not provide a timeline.

The bill remains in pre-legislative scrutiny and is only expected to go before the Oireachtas by September at the earliest.

Transposing the directive will not be straightforward, Kirton said, “as parts of the legislation are technical in nature and present a major change in empowering the National Cyber Security Centre to act in an enforcement role alongside other competent authorities”.

“The Government will need to prioritise the preparation of a bill, which has been promised by the Minister for Justice for later this year, which will no doubt provoke further debate as it proceeds through the legislative process before entering into force,” he added.

Earlier this year, the Commission proposed amendments to simplify NIS2 as part of its digital omnibus overhaul that aims to cut regulatory red tape and make business in the bloc easier.

Amendments to NIS2 aim to increase legal clarity by simplifying jurisdictional rules, streamlining the collection of data on ransomware attacks and facilitating the supervision of cross-border entities, the EU argued.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

We’ve seen a lot of “smart art frames” at CES over the years. Most of them feel like glorified digital photo frames in turtlenecks. However, there’s one that feels genuinely different: Fraimic, and I say that as someone deeply skeptical of this category.

The pitch appears quite compelling at first. Speak a prompt into the device, and its built-in mic sends the command to OpenAI’s GPT Image 2.0, which then generates full-color artwork that lands on a Spectra 6 E Ink display. 

What makes it stand out from competitors like Aura and SwitchBot?

Normally, you’d take out your phone to do that with a regular digital photo frame, but with Fraimic, you just have to tap, speak, and watch something appear on your wall that looks more like paint on paper than pixels on a screen.

The device also features an accelerometer that determines whether it’s oriented in portrait or landscape.

Coming to the competition part, Aura Frames require a subscription and don’t let you swap out the surrounding frame. SwitchBot frames, on the other hand, do not support voice generation. Fraimic does both, while keeping your prompts and images private by default. 

The company offers you 100 free AI generations per year and also provides access to thousands of public-domain works from the Metropolitan Museum of Art. A REST API even opens it up to smart home integration for developers.

So why is the price such a tough pill to swallow?

Because $499 for the 13.3-inch and $1,499 for the 31.5-inch sounds a bit too steep. Aura’s comparable frame runs around the same for the smaller size, but it also offers buyers smaller options that cost even less. Switchbot sells a 31.5-inch variant that costs $200 less. 

It’s worth noting that the 13.3-inch ships now, but the 31.5-inch shows a July 2026 shipping date on the official website. To make the brand’s case, it did grab a Red Dot Award: Product Design 2026. But for a first-gen device from a Chicago startup, it appears to be asking a lot of your wallet.

To me, Fraimic appears to be sitting in an awkward but interesting spot. It’s too expensive to be an impulse buy, too genuinely capable to dismiss. Anyway, we’ll reserve our final verdict for later, when we actually get our hands on it.

SOFTWARE

Linux distro accuses former contributor of deleting years of work and pushing a package that could have broken installs

OpenMandriva has accused a former contributor of using his trusted admin access to trash repositories and push a package that could have broken desktop installations after a community dispute spilled over into the project’s infrastructure.

The Linux distribution disclosed the incident in a forum post this week, describing what it called an attempted act of “distribution sabotage” allegedly involving Davide Beatrici, a developer known for his work on the Mumble instant messaging app.

According to OpenMandriva, Beatrici joined the project some time ago and later offered to migrate its repository infrastructure from GitHub to his privately operated OneDev instance, mirroring several dozen repositories in the process.

While some maintainers were uneasy about concentrating so much of the project’s infrastructure in one person’s hands, the proposal went ahead because, as the project put it, “he was such a well-known figure that we didn’t expect anything bad.”

OpenMandriva says trouble started after two other contributors joined alongside Beatrici. One allegedly engaged in repeated abusive behavior toward users and project members, much of it in private messages. The project says several contributors left before the maintainers finally stepped in, kicking the individual out of the OpenMandriva-Cooker Matrix chat. He wasn’t banned from the project, but OpenMandriva says the decision “triggered a cascade of events.”

Beatrici and another contributor then resigned. When OpenMandriva later decided there was little point continuing to mirror repositories to Beatrici’s private infrastructure, it says it began severing those connections.

According to the project, that didn’t go down well. “This infuriated Davide so much that, abusing of the administrative privileges he still had, he sabotaged the distribution today in the early morning hours,” the statement says.

OpenMandriva alleges Beatrici deleted parts of its GitHub repositories containing years of development work. It also says he “decided to publish an empty package in the cooker repository, which obsoleted all gnome and cosmic packages, which could have damaged the systems of people using gnome or cosmic.”

The Cooker repository is OpenMandriva’s rolling development branch, not a stable release, so the damage appears to have been confined to bleeding-edge users rather than to everyone running the distro. Even so, having one disgruntled admin yank years of work and potentially break package updates isn’t the sort of resilience test most projects volunteer for.

The project says it is restoring the deleted repositories and repairing the affected packages. It also says it carried out “a full system audit” and found that “aside from the removed packages, we found no other violations.”

OpenMandriva adds that it considered legal action, saying the alleged sabotage “constituted a criminal offense,” but ultimately decided against it.

According to tech publication The Lunduke Journal, Beatrici said that “this was by no means a sabotage. The objective was not to harm the distribution I cared for.” He reportedly admitted deleting Cosmic and Gnome repositories and said he did this because someone was “messing with my work.”

The Register has contacted OpenMandriva to ask whether any stable releases were affected, how many repositories were deleted or modified, and what changes the project plans to make to administrative access. We also reached out to Beatrici but have not heard back. 

Every project needs contributors, but they don’t all need the kind of access that can turn a disagreement into a recovery exercise. ®

• The world’s largest digital camera begins recording the changing universe every night
• Giant Chile observatory discovers thousands of hidden asteroids during early testing already
• A new sky survey captures fresh cosmic images every forty seconds overnight

A camera roughly the size of a small car has begun the most ambitious astronomical survey ever attempted from Earth.

The Vera C. Rubin Observatory, perched atop Cerro Pachón in northern Chile, officially started its Legacy Survey of Space and Time.

Bezos is reportedly set to invest $2bn into the company himself.

Jeff Bezos’s Blue Origin is reportedly nearing closing a $10bn funding round that would value the space company at around $130bn. This would be the first time Blue Origin is opening itself up to outside investors since being found in 2000.

Bezos is set to invest $2bn into the company himself, while Coatue Management – which has close ties to Bezos Expeditions – is committing $4bn, reports suggest.

The remaining $4bn has seen significant demand, sources told news publications, mirroring recent investor appetite around major tech IPO listings.

“We finally have enough visibility into our future and our financial success,” Bezos told CNBC in May. “It’s a good time actually to start thinking about the future and bring on some other outside investors.”

The reported funding into Blue Origin comes weeks after its biggest rival, SpaceX, raised a record-breaking $85.7bn in its IPO listing (including the underwriters’ option).

The Elon Musk-owned company has filed for a satellite constellation of up to 1m with major plans for orbital AI data centres. The company currently has more than 10,000 active satellites in orbit.

Earlier this year Blue Origin launched ‘TeraWave’, a new communication network with a planned constellation of nearly 5,500 satellites. The constellation is set to be deployed from Q4 2027, the company said in January. Blue Origin claims that TeraWave will deliver connection speeds of up to 6Tbps anywhere on Earth.

Plans, however, suffered a setback this May after one of Blue Origins’ New Glenn rockets exploded on the launchpad during a hot-fire test. The US Federal Aviation Administration ordered the company to investigate a previous malfunction during a failed launch test in April.

Meanwhile, Bezos’s other big venture, Amazon, purchased satellite telecommunications provider Globalstar to better compete with Starlink.

The $11.6bn deal, announced this April, sees Amazon acquire Globalstar’s existing satellite operations, infrastructure and licences. According to Reuters, Globalstar has 32 planned active low-Earth orbit satellites.

The e-commerce giant plans to integrate Globalstar’s assets into its own space internet service Leo, which aims to have more than 3,200 satellites in space. Currently, the company has more than 375 satellites in space already, and is planning several launches over the course of the year.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

OpenAI unveiled its newest family of models on Thursday, introducing a new set of heavyweight programs into an increasingly crowded field of AI offerings.

GPT-5.6 comes in three variants: Sol (considered its workhorse), Terra (a more intermediate option), and Luna (its budget friendly option). These models expand what users can do across a variety of fields — with the company promising powerful capabilities in enterprise work, coding, and even scientific research.

CEO Sam Altman has promised that his company’s newest models are orders of magnitude more efficient and cost-effective than previous versions, recently telling CNBC that Sol is 54% more token efficient when it comes to AI coding tasks.

Most notably, the company calls 5.6 its “strongest cybersecurity model yet, achieving frontier performance with significantly fewer tokens.”

Indeed, much hubbub has been made about the model’s cyber capabilities, as the Trump administration previously sought to restrict its rollout, ostensibly due to fears of how the model could be misused. GPT-5.6 supports defensive activities, including threat modeling, code review and patching, and blue teaming (simulating an attack on your own systems to find weaknesses before real hackers do).

OpenAI also released a new tool called ChatGPT Work, which — just as it sounds — is designed as a workplace companion for enterprise teams, running on desktop, web, and mobile, that can help with daily clerical tasks, like drafting documents, spreadsheets, and presentations.

OpenAI’s newly announced family of models follows on the heels of similar releases this week from competitors SpaceXAI and Meta.

However, GPT-5.6 and its attendant marketing seems most designed to take aim at OpenAI’s primary opponent, Anthropic. Anthropic has managed to make itself the likable underdog of the AI race, focusing fixedly on enterprise customers and winning a growing share of support as a result.

Not to be outdone, OpenAI cites the Artificial Analysis Coding Agent Index, a notable benchmarking metric, to claim that its latest family of models outshines Anthropic’s models at every turn.

OpenAI calls Sol its “best coding model yet,” and has explicitly compared it to Anthropic’s recently released (and much hyped) Fable. Using the Coding Agent Index, OpenAI claims that Sol “sets a new state of the art at 80, 2.8 points above Fable 5, while using less than half the output tokens, taking less than half the time, and costing about one-third less.”

It adds: “That advantage extends across the family: Terra performs just above Fable 5, while Luna outperforms Opus 4.8.”

The company says that 5.6 is now available across ChatGPT, Codex, and the OpenAI API. Availability per million tokens is priced as follows: Sol is $5 input / $30 output, Terra is $2.50 input / $15 output, and Luna is $1 input / $6 output.