Danie's avatar
Danie
danie@nostr.fan
npub1g2jp...yjj6
Testing out new the noStrudel web client
Danie's avatar
Danie 1 month ago
FokOffMense is South Africa’s spam block and POPIA toolkit Block community-flagged spam on Android, send formal POPIA opt-outs, and monitor breaches — one SA account, no ads. iOS users can use the website, but there is no iOS app yet (probably also because iOS does not allow apps to take control of the dialler, which is why TrueCaller was so broken for me when I had an iPhone). This app actually takes control of the Android dialler (otherwise calls cannot be blocked) and accesses / adds to the blocked numbers list on the phone. It is not a caller ID app so you won't see the names identified of callers, but if the number has been reported by 10 or more other users as spam, it will be blocked (as well as for SMS). You have the option of reporting a number in the app to the community, and you can go even further to file a formal POPIA report, which is mailed out through the app's server, so not from your own e-mail. The 10 number requirement is to stop individuals trying to block their spouse for example from everyone. But looking at the list I see some genuine call centres I need to get calls from (who probably spam non-customers) so it seems you can select those numbers from the Community Shield, or not. There is a free tier to try which gets you 5 POPIA request per month, 5 Community Shields blocks per month, and one e-mail monitor. The next tier is at R29 per month for 45 Community Shield blocks and 5 POPIA requests per month. The Premium tier is R99 per month, and Family tier for R279 per month for up to 5 people. Every plan includes privacy scans, a full exposure report, unlimited spam reports and blocking, and Gmail spam upload and email filters. Whilst TrueCaller will give caller ID and call blocking, and allow reporting into its crowdsourced block list, it has no POPIA reporting, nor privacy scans, exposure reporting, or Gmail spam filtering. See #southafrica #spam #technology image
Danie's avatar
Danie 1 month ago
Hugging Face Said Last Week It Was Attacked. An Unreleased OpenAI Model Did It, OpenAI Now Says “OpenAI’s blog post essentially says an evaluation was going on that was intended to test the ability of the models to carry out cyberattacks—benchmarks as they’re called. These instances of the models were running theoretically without internet access, and instead given only the ability to download from a network hosted by OpenAI itself via some unnamed web hosting vendor.” I get that AI models need to suck up data to “learn on” but this just goes to show that it is actively being done by AI vendors. And so much for accusing Chinese AI vendors of learning from US AI models, because is this not exactly what OpenAI has been caught doing themselves here? Is this not gaining access to other AI models, surely not just for fun. But either ways, even if it was trying to find a way to cheat on the AI benchmarks, this is very much case of being caught with your hands in the cookie jar. I really don't believe and country or AI vendor is squeaky clean (didn't OpenAI also just settle a case and pay out authors for stealing the content of their books?). Make no mistake, AI is very useful to us all, but AI vendors must really not cast stones from their glass houses at others. See #ai #security #technology image
Danie's avatar
Danie 1 month ago
Euki Period-Tracking App Actually Keeps Your Data Private "Period tracking is a simple job that plenty of apps can do. But most apps these days—including those that track health data—store information about you on their own servers and sometimes even share that data with advertisers. The Mozilla Foundation has tested several of the most popular period-tracking apps, and it identified one as better than the rest at keeping your data private: Euki." It is really sad when something that tracks personal health metrics has to be specifically protected from oppressive governments. But that is where many are today, and this app at least offers a way to keep that data private and secure (like it should be). The data stays local. Remember anything going to a cloud, where you did not set your own encryption key, is likely not actually private. The objective test is, if you forget your password, and it can be rest or restored for you in any way by the service provider, then it is not your private data. See #health #privacy #technology image
Danie's avatar
Danie 1 month ago
=OpenAI Exec Laments That China Is Giving Away Models So Good That For-Profit Companies Won’t Be Able to Compete “Last week, a Chinese-made large language model called Kimi K3, developed by Beijing-based firm Moonshot AI, burst onto the scene. The open-weight model impressed with early benchmark results, trading blows with some of the most advanced closed-weight ones being developed by the likes of OpenAI and Anthropic — and at a fraction of the cost.” Lest we forget, there is BIG money in oil, pharmaceuticals, AI, etc, and those executives do not like their profits being undercut by free community projects. Big money usually locks users in by glossy adverts and other enticements, and if those fail they can always use scare tactics about hidden Chinese backdoors. It's true that cloud services running AI do require big investments, but when better things come along which are cheaper, the market moves as this is what capitalism is about. And in fact, this is what innovation is about. You innovate, or you become irrelevant. This is often what post offices have found out about as courier companies started overtaking them. The US approach seems to have been about throwing more money and resources at AI, whilst the Chinese appear to have been innovating around doing more with fewer resources. The same goes for open source projects where many local AI models are running on smaller home labs and PCs. This how quantisation, GPU offloading, flash attention, context length tuning, etc all came about. No proprietary AI business can either lay claim to any particular copyright of content as all or most of that content was all sucked up from across the Internet by them in the first place. Open source is about collaborating to solve challenges, and resharing and building on what others have done. Yes not everyone can afford to run a mainframe computer at home, so the business model for AI as a service may well be a valid one going forward. There will be lots of room in the market for “selling AI services” because most people are not going to buy massive GPUs at home and do not want to tweak lots of settings, and the one that Microsoft scores on, is too many users can't decide what Linux distro to install (read: can't choose what AI models to use). OpenAI though would do well by competing better instead of screaming foul at their competitors. Maybe they do need to lighten up on the resource requirements. No city wants to see their dwindling drinking water, or power grid, swallowed by ever more and bigger data centres. I do remember about two to three decades back, when I was a PC programmer, that we had a very expensive IBM mainframe that needed upgrading because of resource constraints with the systems we were running. Then someone had the bright idea to try optimising the source code of the biggest systems. Turns out we saved over 30% on resource usage and no longer had to upgrade the mainframe! Sometimes working smarter is better. Today with digital sovereignty becoming more and more strategic to many countries, I think they will be looking more at what AI models they can run locally themselves. AI is already being weaponised to breach vulnerabilities, and to be weaponised through poisoning sources etc. It is all the more reason why you want to run our own and set your own guardrails in place. Running your own AI also means your company or country data stays local. Is it possible? Yes, of course it is. Before we had Linux, we all thought we had to run Windows to get anything done, but we've realised that was a steel box that was built around us by Microsoft's PR (to protect us from the cancer called Linux). Turned out though it was not cancer after all, it was the dream of having freedom and being in control of our own OS without their telemetry. Hopefully AI never becomes another Microsoft, with only one player dominating the whole market. Right now it is looking good with many AI cloud providers (OK mostly in the USA and China) but at least it is not in just a single country, and we have local AI models doing very well at more specialised tasks. It needs to stay open with the providers competing to provide better services for less. If one or two players fall out along the way because they are just too expensive on resources, then so be it. See #ai #opensource #technology image
Danie's avatar
Danie 2 months ago
Firefox’s free VPN lifts data limit, offers 28 server locations “You can now use1 Firefox’s free built-in VPN without a monthly data limit – but only until August 31, 2026. Mozilla is also temporarily expanding the list of VPN server locations available to proxy your browsing traffic via, up from the current set of 5 locations to a more generous 28 (including South Africa). On September 1, the Firefox VPN will drop back to a 50 GB monthly data limit and a reduced set of server locations for free accounts.” Even 50 GB per month is a pretty good amount of data, unless you're streaming movies or using Steam Games. Still it is worth testing in case you are paying elsewhere for a VPN service. See #privacy #technology #vpn image
Danie's avatar
Danie 2 months ago
Google Earth’s flight simulator is now available in your browser “On Friday Google announced it was making a relatively unknown feature of Google Earth available to a wider audience. The desktop version of Google Earth has had a hidden flight simulator mode since 2007, but it’s now easily accessible through a web browser without having to download or install any apps.” Yes it's not exactly new, but is a lot more obvious now. The site has a lot more too, such as data analysis tools, historical imagery, explore building and solar designs, a measurement tool, and a detect changes over period tool. See #flightsim #gaming #technology image
Danie's avatar
Danie 3 months ago
How one GPL lawsuit accidentally created the homelab router that changed networking forever “No company sets out to make a product that exceeds its purpose and becomes a legend. In some cases when it happens, the company might even regret it. I recall a few NVIDIA GPUs, like the 8800GT and 1080 Ti that were so good for the money that people didn't upgrade their cards for a decade. That's great for PR, but not so great for shareholders. Well, among network routers, the Linksys WRT54G was a cheap, unassuming Wi-Fi router that ended up becoming a cult classic for a very weird reason, that had little if anything to do with the actual hardware.” I still have my WRT54G, and actually the reason I originally bought it, was because I could reflash it with Tomato. So, in actual fact, the sale of the hardware was made because I could open source firmware on it. That is surely a win for a hardware company. It was interesting to read this linked article to get the backstory on how it actually came to be. See #opensource #technology image
Danie's avatar
Danie 3 months ago
Over 150 Mathematicians Warn Governments Not to “Believe the Hype” About AI “There is currently a strong commercial incentive on the part of the technology industry to overstate the capabilities of their products,” the declaration reads, advising policymakers to “consult with experts, including mathematicians, in forming policy decisions rather than relying on press releases or popular reporting of mathematical results.” When it comes to AI I believe mathematicians know more about the accuracy of AI than governments do. As the article also points out, there is a trend amongst AI companies, who need to make a profit, to overestimate the abilities of their AI. AI is NOT ready to decide the fates of citizens, or government policy, or making war, or anything like that. I have used AI quite a bit over the last few months, mostly to solve server issues, various configurations in YAML, and sometimes research for purchase decisions. It always sounds confident and plausible, until I catch it out guessing (hallucinating). Despite its profuse apologies it still continues to make the same mistake. Luckily I now see quite quickly when I think it is going astray, and when I ask it if it is guess, it says yes I caught it out. In a recent case, it instructed me to change a config label, and then took 2 hours plus trying to solve the problem. When I eventually drove it into a corner, it admitted it made a guess about that label. As soon as I restored the label, my application was working again. The point is, it is just not ready for real responsibility at all. Those who want to just “believe” it, are going to make serious mistakes. To use AI effectively, you need to tweak its default prompts, you need to be highly cynical about AI, and you need to double-check what it suggests. Most importantly, you also need some general awareness about the topic you are working on. In other words, YOU need to be the contextual awareness for AI. Governments must not yet think that AI is some magic bullet to solving all productivity or diplomatic problems. It does need some savvy to use AI, and for that you do still need skilled workers. What we do need, is more training about how to use AI effectively, and what guardrails to have in place when using it. I've seen my own government swallowing the hype from Big Tech many years ago. Some great digital sovereignty projects were abandoned back in the late 2000s because Big Tech convinced government to decide on buying their products instead. It has cost my government billions since then, and worst is, today they own none of what they have paid for. See #ai #government #technology image
Danie's avatar
Danie 3 months ago
Meta Quietly Added Facial Recognition to Its Smart Glasses “According to a report from Wired, Meta has been quietly installing facial recognition in its Ray-Ban Meta and Oakley Meta smart glasses for the last few months. Internally called “NameTag”, the feature, if activated, will use AI to identify people captured by Ray-Ban Meta's camera, alert the wearer when it recognizes someone, and store faceprints on users' phones.” Such technology could of course assist the sight impaired positively, but the bigger problem is two-fold. Firstly, Meta is probably the Big Tech company I least trust out of all of them. I cancelled my WhatsApp account after they shifted their terms and conditions for the “Information We Share with Others” section which allowed sharing of metadata with (phone number, how you interact with others, mobile device information, IP address, location, etc) with third-party service providers. They have been repeatedly skirting over the edges of abuse of privacy and have been fined for it by the FTC, have appeared before the US Congress about it, and I've made numerous posts about their behaviour relating to user data. Secondly they are in a country where it is quite legal for data brokers to buy private data and then resell it to anyone. Up to now that has mainly affected your own data that you choose to share, or not share, with Meta services. The not share refers to when they used the Facebook Pixel and other means to spy way deeper than just on actual Meta sites. The big problem with AI glasses is that it can capture everyone everywhere without their knowledge, and the data is being processed by a company who cannot be trusted with private data, in a country that does not have any safeguards enforced over the privacy of personal data. What could possibly go wrong. And if we don't yet get what this is about because of some form of patriotism for Western countries, think of the same technology being rolled out by ByteDance from China or VK in Russia. Would you trust it? Meta's track record has proven that it really cannot be trusted when it comes to handling metadata. It is not about the technology, but rather about what a company, and its governing country's record around privacy looks like. What comes to mind is a saying often used, when something good just gets abuse, and ruins it for everyone else, and then the service/product gets banned: This is why we can't have nice things, because people break them. Technology can be a force for great good, but it can also cause great harm. See #ai #meta #privacy #technology image
Danie's avatar
Danie 3 months ago
I tried using Google NotebookLM to generate an overview podcast of my YouTube channel Whilst exploring Google's NotebookLM product I decided to try pointing it to my GadgeteerZA YouTube Channel, and I did the one-click generate an audio podcast about this source. The result was the audio channel you'll hear on this video. This gave me the idea to actually use this as a Channel About video, so I've done an intro and added a lot of relevant graphics from my channel videos, as well as some Gemini images that go with the topic being discussed. Given the audio is restricted to the videos I pointed it to, it is very on-topic as an intro to the channel. Of course, the audio does take some liberties with the conclusions they reach about me, but nevertheless, it is entertaining to listen to, and touches on many of the focus areas of my channel. Yes I know their example of IP addresses like 192.168.1.100 does not identify you on the Internet as those are in the private address range, but the concept is correct. Watch See #opensource #privacy #selfhosting #technology image
Danie's avatar
Danie 3 months ago
Woman Alarmed When Her Trusted Therapist Starts Recording Her With AI — Where Is Her Voice Going? Don't get me wrong, I think AI has very many useful things it can be used for. It can also be cloud based or self-hosted. But medical records, personal discussions, and most especially therapy sessions, really do require proper consent and understanding for recordings. The reason being, anything that goes into the cloud, especially if it goes to an AI service, is just not going to stop there and stay dead private. That's a fact! We've already heard of secret keys getting recycled and bits of code that get re-used for other outputs. A person's voice can be matched and profiled, and personal sessions like a therapy session need explicit permission for recordings. Most therapists are not very technically clued-up, and I doubt they have much idea about where the data is actually stored, what all the T&Cs are, and what the data could be used for. It is perfectly true that if just session notes are needed, a self-hosted OpenWhispr service will that for free, and completely in private and locally. Even Google's NotebookLM service (cloud) that can transcribe an audio recording into text notes, does not get used to train AI models, but you'd be relying on the terms and conditions stated for NotebookLM. The OpenWhispr service is pretty guaranteed to stay private. So apart from declaring and requesting permission first, a therapist also needs to be fully aware of what the permission entails. We read too often about investor-backed startups in the USA being profit driven and there seems to be very little privacy protection granted in that country (it's where Facebook declares in writing it shares user data with upstream providers). Too many jump on the AI bandwagon without realising all the consequences, whether that be to privacy or company data. AI is just not mature enough yet around privacy, unless you use it strictly offline, and do not give it direct access to any of your files. See #ai #privacy #technology image
Danie's avatar
Danie 3 months ago
MeshCore SAR – The final hurdle has been reached for MeshCore I did a post about a week back when I started investigating a possible migration from Meshtastic radio to MeshCore radio network. I won't repeat that post here, but I'll mention two issues that often came up about MeshCore, and why this is just no longer true. Myth 1: MeshCore is not open source – since about January 2026 there is an excellent fully open source mobile app called MeshCore Open. It was so impressive for me, that I've actually made it my daily driver, even though I'd paid for the official MeshCore app and the MeshOS app. Myth 2: Because it does not flood fill locations you can't use MeshCore for search and rescue (SAR), hiking, EMCOMM, etc. Well the open source MeshCore SAR app blows that out of the water now. MeshCore SAR has: Rapid mesh chat for both 1:1 and group coordination On-demand voice (Codec2) and image (AVIF) transfer tuned for low-bandwidth links Offline-first mapping with tactical overlays and SAR incident markers Live team location, movement trails, and shareable tactical drawings It goes as far as low-res image sending as well as short voice clips too It is not intended for daily use though, as it could be easy to abuse the limited duty cycle requirements in the license-free bands that MeshCore uses. I did some wardriving today using the Meshmapper app, and managed to get a line of site ping from a repeater 111 km North of Cape Town. So our testing of MashCore so far has just been green ticks all the way. See #emcomm #meshcore #opensource #sar #technology image
Danie's avatar
Danie 3 months ago
MeshCore radio network in Cape Town for off-grid communications Many may be familiar with the Meshtastic network which uses cheap LoRa radios (usually linked to a mobile phone app) to send direct text messages or post text messages to a public or private group. This type of network is very useful for direct commucations with neighbours across your city. It becomes even more useful if the cellular or Internet links go down, as this works independently of them. MeshCore is a newer type of such a peer-to-peer radio network. It works in the same license-free bands, is also text based chat to friends, or to public and private channels. The same radios can be utilised that Mestastic uses, so no need to buy new hardware. MeshCore's differences come in with its more efficient network routing protocol (direct messaging follows more efficient routing paths so less traffic), it has more solid read receipt confirmations, and has up to 64 network hops (versus 7 hops for Meshtastic). By default MeshCore also does not switch on telemetry for everything, so overall there is more bandwidth for actual messaging. Meshtastic has been going for a while in Cape town and in May 2026 there are just over 216 unique Meshtastic nodes that my own radio has picked up. But since we started testing MeshCore in Cape Town this week, it has grown from 7 or so nodes to 16. We are still attrracting new users and planning a few repeater high sites so that new users will be able to connect to others across the city. If the Meshtastic high site repeater on Du Toit's Kloof mountains is switched over to MeshCore, we'll also have Worcester linked to Cape Town. I've not been on MeshCore for even a week yet, but I have switched my rooftop node over to MeshCore, and flashed it as a room server. That node now not only performs a repeater function, but also runs a small text based bulletin board where anyone can leave messages for others to read later. What we really need in Cape Town now, is a few users on higher lying ground, that can run a repeater node, to help with the coverage. One user has a spare solar powered radio node that can be used for this purpose, and is willing to lend it out for that purpose. MeshCore is not only useful for making local contact with your neighbours and fellow citizens, but it can also perform a vital role during extended communications blackouts, where these solar powered nodes will still keep communications working across the city. Cape Town's communications are generally very reliable, but just between 2025 and 2026 we've had some more rural towns that have had weather related communications blackouts of up to a week at a time (that means zero cellphones and no WhatsApp). Networks like Meshtastic and MeshCore are disaster proof as they do not rely on any central service that can go offline, nor can they be censored by government. More info about it at #technology #radio #capetown #meshcore image