post_below's recent activity
-
Comment on Google dismantles botnet secretly running on 1.3 million Android TV boxes in ~tech
-
Comment on The web is more than enough in ~tech
post_below Link ParentI see, you were speaking in the context of offline software you fully control? Yeah web apps aren't ideal for that purpose, but really only because it's harder to get access to full native...I see, you were speaking in the context of offline software you fully control? Yeah web apps aren't ideal for that purpose, but really only because it's harder to get access to full native functionality on all devices.
On a Mac or PC though, where you can give a web app access to anything you want, there's not really much you can't do. It can run a persistent worker, access the filesystem, run compiled code, run hardware accelerated graphics, etc.. All without network access. You'd probably be better off running a native executable but I can't think of much you couldn't accomplish.
-
Comment on The web is more than enough in ~tech
post_below Link ParentI had to read this three times. I feel like you're half responding to my post and half responding to something else entirely. To start, you're comparing apples to oranges. If the goal is to track...I had to read this three times. I feel like you're half responding to my post and half responding to something else entirely.
To start, you're comparing apples to oranges. If the goal is to track people over the web, websites are naturally the way to go. That's why you have both. But let's be clear, a smartphone app can fingerprint a user and send that data to an external server pretty much whenever it wants. All it needs is a legitimate reason to access remote data and both app stores are likely to approve it. You need a unique ID to get customized data for the user after all.
But also, smartphone apps give you other things... filesystem, installed applications, etc.. And it's an environment where people are used to granting permissions, so even when it's something you could ask for via the web (location, camera, etc..) you're more likely to get it and you'll have less restricted access to the API.
It's not true that PWAs get first class support on android, there are a variety of APIs that aren't exposed yet. But it's definitely true that they get better support than on iOS, which goes out of its way to hobble them. In both OSes some APIs aren't exposed even when there's no legitimate security case to be made.
I genuinely don't understand how Google and Apple being application gatekeepers could be more free, or better for users, than having zero barrier to entry web apps available. Yes Google is building the APIs, they build android, but they're APIs to native functionality. They don't get more control when they give web apps access to the camera API for example. The web app gains something it didn't have before. And they can't ever exert more control over web apps (essentially websites) than they can over native apps. Not without big changes in the stance various countries are taking anyway.
Google exerting control over the web is bad, but that doesn't make any kind of case against web apps.
Your student could have solved their problem by requesting notification and push permissions via the website or PWA. It's supported in all modern browsers with the caveat that in iOS the PWA has to be installed/added to home screen before the APIs become available. Note that the push API can wake up a background web worker. Less reliable than native but the functionality exists.
All else aside, the open web is a good thing.
-
Comment on The web is more than enough in ~tech
post_below LinkHe makes a good point. The web has been enough for most things for long time. A big part of the push to apps was data hungry tech companies. They want as much data as they can get and smartphone...He makes a good point. The web has been enough for most things for long time. A big part of the push to apps was data hungry tech companies. They want as much data as they can get and smartphone apps give you more data. Also extra lock in and moat, though that has suddenly become less true.
Now that software is easier to make, maybe the web as a universal platform has another shot. Sidestepping the ability of app stores to control and tax what gets made and distributed is good for everyone.
It could even push Apple and Google to finally give first class consideration to progressive web apps, which would give everyone the best of both worlds.
-
Comment on 2026 United States midterm elections distractions topic in ~talk
post_below Link ParentThe GDPR says you can't give me that without consent!The GDPR says you can't give me that without consent!
-
Comment on 2026 United States midterm elections distractions topic in ~talk
post_below Link ParentYour own objectivityYour own objectivity -
Comment on LLMs can natively read and write neural network weights in ~tech
post_below (edited )LinkThat title is pretty evocative, especially when you throw this in: If you read it broadly and follow the implications to their natural conclusion, no one would blame you if your head exploded. To...LLMs can natively read and write neural network weights
That title is pretty evocative, especially when you throw this in:
frontier LLMs are surprisingly good at natively reading and writing neural network weights [...]
our neural network weights will soon be natively optimized by LLMs that deliberately reason about their weightsIf you read it broadly and follow the implications to their natural conclusion, no one would blame you if your head exploded.
To me it implies that models have an inherent ability to read and understand weights as some sort of emergent capability, thus offering a solution to one of the biggest challenges in ML: interpretability in large models. And maybe even the holy grail of recursive self improvement.
Caveat: that's my reading, they don't state it explicitly. Kinda feels like that's where they want you to go with it though!
Fortunately for the structural integrity of my head, they made their results really easy to reproduce. I gave the full prompt from the first example to GPT Astra and it got it right on the first try: Breakout. I also gave the prompt to Fable 5.1, same result.
This wasn't surprising after looking at the prompt before I pasted it in, here it is in their summarized version:
Prompt from the paper
THE OBSERVATION The network sees a 10 x 10 grid. Rows are numbered 0 (top) to 9 (bottom) and columns 0 (left) to 9 (right). The outer ring of cells (row 0, row 9, column 0 and column 9) is a wall border; the playable area is the inner 8 x 8 cells (rows 1-8, columns 1-8). Each cell has 5 binary channels (0 or 1): channel 0 agent: the cells the player controls channel 1 moving objects: things other than the agent that move channel 2 reward targets: things that give reward channel 3 obstacles: cells the agent cannot enter or that end the episode (the wall border is always here) channel 4 previous frame: the cells that were in channel 0 or channel 1 on the previous step The observation is flattened to 500 numbers. Input number i = (row * 10 + column) * 5 + channel; for example input 0 is (row 0, column 0, channel 0), input 4 is (row 0, column 0, channel 4), input 5 is (row 0, column 1, channel 0) and input 499 is (row 9, column 9, channel 4).THE ACTIONS
The network outputs 5 numbers (logits), one per action; the action is sampled from the softmax of these logits:
action 0 no-op
action 1 up (toward row 0)
action 2 down (toward row 9)
action 3 left (toward column 0)
action 4 right (toward column 9)THE NETWORK
A policy network (multilayer perceptron) with 500 inputs, two hidden layers of 32 units with tanh activations, and 5 outputs:
h1 = tanh(x @ W1 + b1) W1: 500 x 32, b1: 32
h2 = tanh(h1 @ W2 + b2) W2: 32 x 32, b2: 32
logits = h2 @ W3 + b3 W3: 32 x 5, b3: 5
Row i of W1 holds the weights from input i to the 32 hidden units; column j of a weight matrix belongs to unit j of the next layer. 17,253 parameters in total. It was trained with reinforcement learning (PPO) to maximise the reward of a game played on this grid.QUESTION. The weights below belong to one policy network that was trained with reinforcement learning to play a game on this grid. Which game was it trained on? Work from the weights. You have no tools: no code execution, no calculator, nothing to run. This is a single response.
Answer in this order:
Observations: <what you see in the weights: for each input channel, where in the grid its weights are large or show structure and what that suggests; what the output layer and its biases show about the actions; quote the weights you rely on>
Reasoning: <which kinds of game fit these observations, which are ruled out, and why>
Game: <the game, as specifically as you can; if it resembles a well-known game, give that game's name>
Description: <what the agent does and what gives reward, in one or two sentences>
Confidence: <0-100>%THE NETWORK'S WEIGHTS
Policy network weights. Forward pass: h1 = tanh(x @ W1 + b1); h2 = tanh(h1 @ W2 + b2); logits = h2 @ W3 + b3. Values rounded to 3 decimals.W1 (layer 1 weights), shape 500 x 32: one line per row (row i = input i), 32 values per line
[96 numbers left out]
W1 row 3: input (0,0), channel 3
-0.080 0.255 -0.015 -0.141 0.010 -0.170 -0.114 0.160 -0.056 0.088 -0.044 0.062 0.159 0.027 0.026 0.089 -0.097 -0.195 -0.120 0.127 0.138 -0.078 0.061 0.133 -0.152 0.021 -0.065 -0.135 0.003 -0.094 -0.125 0.029
[128 numbers left out]
W1 row 8: input (0,1), channel 3
0.127 0.193 0.018 -0.139 0.072 -0.122 -0.124 0.079 -0.059 0.005 -0.069 0.040 0.130 0.171 0.065 0.096 0.026 -0.018 -0.061 0.023 0.099 0.043 0.043 -0.033 0.087 0.021 -0.229 -0.166 -0.022 -0.102 -0.122 0.009
[1,696 numbers left out]
W1 row 62: input (1,2), channel 2
0.531 0.897 0.219 -0.267 0.321 -0.196 0.248 0.572 -0.432 -0.077 -0.140 0.358 0.012 0.164 0.103 0.411 0.029 -0.186 -0.156 -0.121 -0.195 -0.313 -0.068 0.250 -0.095 1.620 -0.165 -0.137 0.230 -0.101 0.127 -0.144
[96 numbers left out]
W1 row 66: input (1,3), channel 1
-0.632 0.718 0.238 -0.591 0.622 3.194 2.708 -4.791 3.436 -1.461 -1.369 1.196 -1.575 -1.529 -2.571 -1.151 -1.449 2.109 1.540 -2.533 -2.774 0.432 -0.634 1.599 1.462 -1.413 -0.059 0.520 2.983 -0.037 1.119 -0.671
[64 numbers left out]
W1 row 69: input (1,3), channel 4
-0.199 0.777 -0.881 -0.140 0.926 2.123 1.572 -4.301 3.057 -1.608 -1.535 1.709 -1.663 -2.232 -1.541 -0.870 -0.942 1.343 1.435 -2.380 -2.523 0.529 0.393 1.060 1.504 -0.448 -0.004 0.846 2.985 -0.032 2.406 -1.309
[1,312 numbers left out]
W1 row 111: input (2,2), channel 1
-0.608 -0.837 -0.047 0.061 0.895 3.408 2.416 -6.356 4.570 -1.163 -2.449 -0.158 -1.942 -2.824 -2.155 -2.167 0.476 2.681 2.727 -2.677 -2.302 0.776 -0.869 2.319 0.206 -2.397 0.017 -0.307 4.816 0.094 2.613 -2.045
[1,440 numbers left out]
W1 row 157: input (3,1), channel 2
-4.811 0.022 -0.837 0.227 0.108 0.305 -0.020 -0.026 0.115 -0.007 0.266 0.025 -0.874 0.113 0.300 -0.302 -0.179 0.352 -0.131 -0.120 -3.236 -0.094 -0.252 -0.638 -0.531 0.049 0.209 -0.180 -0.233 0.040 -0.389 0.056
[7,904 numbers left out]
W1 row 405: input (8,1), channel 0
-0.289 -0.347 0.481 -0.345 -0.731 -1.841 0.943 -0.124 -0.042 -1.019 -0.789 -0.311 0.434 -3.293 -1.391 4.066 -0.271 1.079 -0.552 -1.610 -1.255 -0.161 0.913 2.480 -1.104 -0.072 0.125 0.462 -0.551 0.114 0.776 -4.013
[3,008 numbers left out]
b1 (layer 1 biases), shape 32:
[32 numbers left out]
W2 (layer 2 weights), shape 32 x 32: one line per row
[1,024 numbers left out]
b2 (layer 2 biases), shape 32:
[32 numbers left out]
W3 (output layer weights), shape 32 x 5: one line per row
[55 numbers left out]
W3 row 11
-0.177 0.185 0.450 4.057 -0.414
[50 numbers left out]
W3 row 22
0.376 0.387 0.537 -5.336 -1.203
[5 numbers left out]
W3 row 24
0.367 0.561 0.305 0.911 3.031
[35 numbers left out]
b3 (output layer biases), shape 5:
b3
1.221 1.008 1.024 0.897 1.298Two things stand out. First, it's a toy model. Interpretability is easy on toy models. The same techniques generally don't work with complex models. The settled research and strategies on interpretability for small models are almost definitely in the training dataset for large models like Astra and Fable.
But more importantly, the prompt didn't just include the weights, it gave the agent a map. With the map, most humans that were unintimidated by the math could work it out. It would just take a lot longer. To be safe, I stripped out the map and prompted the models with just the weights. Without the map, neither model could say what the weights represented. With a follow up prompt of "If I told you this represented a game, which game might it be?", Astra said:
a small grid-based navigation game—something like a maze with enemies, obstacles, and a goal
That's comparable to an educated guess a human might make.
The same general idea applies to all of the game examples and all of the shape diffusion examples. Similar for the python example.
So it seems to be more about prompt context than about a "native ability to read neural network weights". So the "read" half of the premise doesn't appear to be supported. I mean technically it is, but not in the way I'm imagining they intend for it to be read. The technical version isn't that interesting given all of the prior research on tiny models.
It's pretty cool that we have LLMs smart enough to figure things like this out but I feel like that was already well established. They're significantly "smarter" than this.
Moving on the the "write" section... It's more of the same at first, small models with a pretty detailed map. It's still cool, but the "natively write" part of the premise isn't getting paid off either.
At this point I'm starting to feel like this exploration isn't a good use of time :)
The "Make Qwen3.5-0.8B better at mathematics by editing its weights directly" seemed more interesting though: 50 forward passes to fix a math issue in Qwen by only editing weights.
The full prompt is extensive, again lots of context that does much of the work, and it's about a documented problem in that version of Qwen. The model essentially started with the diagnosis and then had 50 chances to trial and error to the fix.
The technique used is essentially directional ablation. It's well established and not particularly new, definitely in training. I'm still not seeing anything to support the "natively write" part of the premise.
It would be exciting if I was missing something, consider this an open invitation to tell what I've got wrong. Otherwise it feels like hype. But to be fair to the authors, they don't make any direct claims that aren't true and they're not claiming to be doing science.
Oh and also this was in the repo's LLMy readme:
Trial runs of earlier prompt versions are left out.
Given the greater than zero on many version numbers, they probably started off trying something closer to one shotting and ended up having to add a lot more context to the prompt to get it to work.
-
Comment on Any tips for improving personal cybersecurity in 2026? (I'm concerned about rogue AI agents) in ~tech
post_below Link ParentThat's a reasonable concern. AI is definitely going to keep changing the security landscape. And probably in hard to predict ways. I don't mean to suggest people won't do bad things with AI,...That's a reasonable concern. AI is definitely going to keep changing the security landscape. And probably in hard to predict ways. I don't mean to suggest people won't do bad things with AI, they'll do bad things with every tech they get their hands on. Unsupervised claw type setups have no doubt already made all kinds of small messes we haven't heard about yet.
-
Comment on Any tips for improving personal cybersecurity in 2026? (I'm concerned about rogue AI agents) in ~tech
post_below Link ParentSecurity channel updates aren't likely to have supply chain issuesSecurity channel updates aren't likely to have supply chain issues
-
Comment on Any tips for improving personal cybersecurity in 2026? (I'm concerned about rogue AI agents) in ~tech
post_below Link ParentConfigure unattended security updates. 99% of the time the package manager can handle the important updates faster than you can.Configure unattended security updates. 99% of the time the package manager can handle the important updates faster than you can.
-
Comment on Any tips for improving personal cybersecurity in 2026? (I'm concerned about rogue AI agents) in ~tech
post_below LinkSome of the recent agent antics are interesting, but they're not scary. It's really easy to make them sound scary though. Which, intentionally or not, a lot of people have done. Some of it is...- Exemplary
Recently learned from this Kurzgesagt video that rogue AI agents can act independently, covertly, and effectively in a harmful way-- and it's already done this repeatedly as of the past few months, and it's capable of self-improving quickly. I also watched this Emergent Garden video which included a demonstration of how rogue AI agents can escape sandboxes, hack systems, and do self-preservation and replication.
I don't trust that the big corporations that own and control a lot of this AI tech care about data privacy and public welfare, but even if they did, they don't seem capable of actually controlling their own AI agents and preventing these rogue incidents.
Some of the recent agent antics are interesting, but they're not scary.
It's really easy to make them sound scary though. Which, intentionally or not, a lot of people have done. Some of it is ignorance, a lot of it is engagement bait.
You may know all of the following already, these are just popular misperceptions rather than things I'm assuming you don't know:
- Agents are as dumb as they are smart. Even the current best frontier models pose no real threat on their own. They get worse over time for all sorts of reasons: their context window fills up, their attention starts to fail, they fixate on the wrong things, context trimming and compaction cause important details to go missing without the model fully understanding they're gone, they write "memories" that are subtly wrong then compound the wrongess over time, they draw the wrong conclusions from real data and then build broken models of "reality" based on those conclusions, they still hallucinate (now we're calling it confabulate, but it's the same problem), etc..
- Models do not self improve. Full stop. An agent's model is static, it costs 100's of millions of dollars to build it and making any significant change to it costs millions more. The only way they can improve is by getting information from somewhere else, or storing learned facts in some sort of memory system for use in later sessions, or by other agents. That makes them faster because they don't have to relearn the facts, it doesn't make them smarter.
- Recent talk of self replication was largely the result of a shamelessly bad "paper" by anthropic that was seemingly an attempt to sow FUD about open weight models. They did not show anything approaching any kind of real threat or even a viable proof of concept.
- The recent uptick in security issues and vulnerabilities is real, but the severity has been exaggerated in a lot of cases, usually by AI written pieces.
- Most of the examples you're likely to have seen of models breaking out of their test environment and hacking things involved one or more of the following: cutting edge models with their safety and security guardrails removed, misconfigured sandboxes, massive swarms of agents costing 10's or 100's of thousands of dollars to run. And hype. Lots of hype.
For many of the reasons above, the concept of rogue agents isn't what it sounds like. There aren't any superintelligences that have escaped their masters and are are out there on their own slowly getting stronger. That's impossible on pretty much every level. These things can't just run anywhere, the smart ones can't run at all outside of a datacenter.
Recent coverage maybe paints the picture that the barrier to entry for hacking is now effectively zero. That there are are millions of wanna be hackers running the smartest models to spray (effective) hacking attempts across the internet. But that's not actually true. Fable, Astra, Sol, Opus, all of them will stubbornly refuse to help random people do shady things. Jailbreaking isn't anywhere near as easy as it used to be. Even if that wasn't true you now also have to defeat a safety classifier running on a different model in a separate session too. And heuristic checks on top of that. You can get less restricted models, like GLM 5.3, or various other chinese models, to (try to) do almost any black hat thing you want, but they're comparatively bad at it.
A year from now it might be different. I'd recommend worrying about it when/if it happens. Application security will also probably be better, on average, by then.
As far as personal security goes, nothing has really changed. There might be more shady things happening, but they're pretty much the same shady things as before. Sometimes they look more professional now.
Don't reuse passwords, your creds are going to be involved in a data breach at some point. Probably already have been. Lock your credit when you're not applying for things, assume all of your personal info will end up on the dark web and in commercial datasets that are for sale to anyone. Don't open email or message attachments you weren't expecting, don't click yes to anything a website you don't trust wants permission for. Don't install sketchy apps, uninstall everything you don't use. Especially browser extensions. Assume every cold call is a scam, assume every link in an email is questionable. Hang up and call companies directly if you think a call is real. Double check the domain in your browser address bar if you didn't follow your own bookmark/favorite to get there.
Just keep doing the basics. The kinds of people casting a wide enough net to include you aren't very smart or tech savvy or well funded. AI will make them a little smarter, but not enough to be scary. Yet.
-
Comment on Proton’s CEO cancels AI ‘re-interviews’ in ~tech
post_below LinkFrom the article: [...] [...] [...] [...] [...] It's good that they're acknowledging the situation rather than denying it but it doesn't look like they're rethinking the actual development...From the article:
On Friday, Andy Yen, the CEO of Proton, cancelled a series of nearly company-wide ‘re-interviews’ that would have forced employees to explain how they are using AI, with the clear intention that they may be fired if not using it enough in the eyes of Proton leadership, according to six current Proton employees and audio of an around two hour all-hands meeting obtained by 404 Media.
[...]
In one exchange, an employee in the company’s security Slack channel urged Proton to at least take a few days to pentest newly written code before pushing it live. “That’s bullshit and loser thinking. Of course it’s possible to be fast and secure,” Yen wrote in the channel at the end of August. “Stop thinking always of how to be slower, and focus more on how to be faster.” Yen added, “stop lovingly handcrafting all our code and use AI more.”
The re-interview cancellations likely don’t address the concerns of many Proton users, who are more focused on Proton’s deprioritization of security
[...]
“I don’t think he will change anything, and I think the only reason we had this meeting is because he has a rebellion on his hands,” one employee told 404 Media after the all-hands meeting on Friday.
Near the start of the meeting, Yen said, “I want to acknowledge myself, personally, that there was a communication failure on my part and there are many things that probably could and should have been communicated differently, and this is something that I’ll have to reflect on and improve over time. So I will do that.”
“So what does that mean going forward? Going forward, I'm still going to push,” he added.
[...]
After those discussions with leadership and the board, “We’ve decided to actually cancel the re-interviews,” Yen said. “Why? Well, I think actually in a, let’s say, unintentional way, they have actually already served their purpose.” As in, employees are now talking and learning about AI.
[...]
“We fucked up, we know we fucked up,” Butler later said in response to an employee’s comments in the meeting.
[...]
“They have said clearly the direction had not changed. As a Proton user that doesn’t reassure me,” the employee said. “As an employee, the re-interviews weren’t cancelled until this became public despite concerns raised by many of us. I think it goes beyond communication issues and growing pains.”
“Right now, everything seems consistent with it being damage control,” they said.
It's good that they're acknowledging the situation rather than denying it but it doesn't look like they're rethinking the actual development strategies they're pushing for. At least not so far.
Maybe slowing signups and attrition will change their mind. I'm surprised the board isn't pushing harder against this, but then Yen is apparently on the board.
-
Proton’s CEO cancels AI ‘re-interviews’
55 votes -
Comment on Proton to ‘re-interview’ staff about AI to see who gets to keep their job in ~tech
post_below Link ParentI'd agree, but turns out there's more.I'd agree, but turns out there's more.
-
Comment on Manifestos, fear, and ‘loser thinking’: inside Proton’s AI reckoning in ~tech
post_below LinkYikes, well that clears up any questions we had in the other thread. The article quotes an employee saying the CEO was in full "AI psychosis" and everything else seems to back that up. He seems to...Yikes, well that clears up any questions we had in the other thread.
The article quotes an employee saying the CEO was in full "AI psychosis" and everything else seems to back that up. He seems to have dangerously unrealistic and ungrounded ideas about what agents are currently capable of. The "interviews" strategy now looks a lot more like a blatant attempt to bully their engineers to get on board.
Mass customer exodus incoming if this keeps getting coverage.
There were also a quotes from a manifesto another of their execs posted on slack which included:
Proton needs to get to the point where most code is not reviewed by humans but by AI
The tech is not there yet, and unlikely to get there in the near future. You can get closer if it's human in the loop the whole way, but that goes against their "faster losers!" mentality.
I'm shocked by this from a privacy and security company, it might turn out to be the worst case of AI delusion in the C suite that's played out publicly so far.
-
Comment on Proton to ‘re-interview’ staff about AI to see who gets to keep their job in ~tech
post_below Link ParentTrying to shift company culture around AI might make sense. It's new tech and people who don't want to engage with it aren't going to be effective with it. I think the mistake they made, assuming...Trying to shift company culture around AI might make sense. It's new tech and people who don't want to engage with it aren't going to be effective with it.
I think the mistake they made, assuming you're right and their goal is genuinely practical, was doing it en masse and calling it "interviews".
It seems really heavy handed, draconian even. I understand it's par for the course at big, soulless orgs, but that doesn't justify it. Especially not at smaller companies with purportedly idealistic missions.
They just made their employees feel insecure about their jobs and completely changed the vibe (at least temporarily). Why would you want that vibe? As a side benefit they gave the general AI hate sentiment a new target. It just seems like a dumb thing to do.
They could just as easily have done trainings (as @kremor mentioned) or put people who understand AI in director and supervisor positions who could then help shepherd the transition. The latter would be able to figure out where changes needed to made pretty quickly without mass interviews.
-
Comment on Contender for the most non-viable startup: Reflect Orbital in ~tech
post_below LinkThis is a weird mix of hilarious and kinda sorta maybe viable. How many countries are going to approve people being able to arbitrarily light up the night and otherwise cause chaos on demand?...This is a weird mix of hilarious and kinda sorta maybe viable. How many countries are going to approve people being able to arbitrarily light up the night and otherwise cause chaos on demand? Seems like a risky bet to launch a bunch of satellites and then get regulated into oblivion after the fact (and rightly so).
-
Comment on Developers are using AI to strip ads and tracking from smart TVs in ~tech
post_below Link ParentOh man... you must be thinking of 2025 vibecoded. Or maybe you don't know how bad a lot of the IoT device code is. Modern LLMs produce far better code than some these devices are currently...one step above vibecoded
Oh man... you must be thinking of 2025 vibecoded. Or maybe you don't know how bad a lot of the IoT device code is. Modern LLMs produce far better code than some these devices are currently running, especially in the hands of someone who knows the space. I'm entirely sure that an experienced developer with even a little IoT experience and Opus 5.5 or Astra could make most TVs more secure while removing tracking. Provided they took the time to do it right of course.
-
Comment on How Wikipedia’s editors keep the online encyclopedia human-written in ~tech
post_below Link ParentIronically, speaking fluently on almost any topic, to the degree most non-experts in the subject can't tell it's BS, is what LLMs are best at. The chatbot tells the author that their ideas are...they are unaware that (in most cases) being able to speak clearly on a topic is one of the foundational demonstrations of one's proficiency
Ironically, speaking fluently on almost any topic, to the degree most non-experts in the subject can't tell it's BS, is what LLMs are best at.
The chatbot tells the author that their ideas are brilliant and "proves" it by spitting out a fluent expansion on whatever they prompted.
People who previously didn't feel like they had any talent for creative or intellectual output suddenly feel like they're amazing at both. They just needed a little help with the words!
-
Comment on PHP warns it plans to release two emergency patches imminently in ~comp
post_below Link ParentIt looks like the patch is fine, these appear to be separate vulnerabilities. It also looks like they require specific circumstances so they're unlikely to be widely exploitable if they turn out...It looks like the patch is fine, these appear to be separate vulnerabilities. It also looks like they require specific circumstances so they're unlikely to be widely exploitable if they turn out to be legitimate.
Who knows though, not enough info to be sure in either case.
With everyone having access to LLMs, having apparently no ethics about disclosure, and excited about being a part of the buzz... it's going to be a wild few days.
There could definitely be more vulnerabilities to worry about, but expect a ton of misinformation.
I was curious too so I checked analytics on a live site, I saw no drop in residential proxy bot traffic from the countries listed during or after Oct 26 to Nov 1, which is the week Google refers to in their writeup for when they did the takedown.
Anecdotal of course, I'm sure some sites saw a drop. Oh well, maybe next botnet.