I feel like Meta are laser-focussed on gimmicks to try and be the next big thing.
Animated avatars do not a killer app make though. They need to focus on usability and function not this crap.
335 karma · joined June 5, 2017
Animated avatars do not a killer app make though. They need to focus on usability and function not this crap.
Consumers are option-rich. AI companies are backed into a corner.
Right now inference costs a fraction of subscription prices. For AI companies that are making losses - like OpenAI - that's R&D costs and infra driving it. Not inference.
If OpenAI up their pricing now that opens the market for open weights models to eat up marketshare. OpenAI might stay relatively dominant but if a smaller company can profitably offer a frontier open model (that they didn't r&d) on a smaller server farm at $20/month profitably, then the shock is vastly limited.
And that's not the only solution. Hardware costs are sky-high now, but capacity is growing and innovation is being fuelled. If AI companies do raise prices some may see enough move to the cheaper alternatives for them to crash and burn. That will be a bubble deflating or bursting and it will flood the hardware market with cheaper RAM and other components. That's when those open models become viable for running at home, private and personal. Something a lot of people will find appealing. Then the cost to run becomes your electricity bill, or even solar powered.
US grid capacity is said to be a fraction of what it needs to be in coming years. Expansion, and green energy expansion across the world, as well as rising oil prices and instability mean electricity is only heading downwards in cost.
A shock is coming, but it's not going to be felt by everyone. Nowhere near.
One problem I have with a lot of this 'AI isn't sentient' stuff is - who is saying it is? Is their opinion of any value? I see AI companies make marketing claims about AGI and the singularity. But their opinion is without value in this debate, because of their bias. No big name in AI academia or philosophy of mind seems to have made a strong claim about sentience in AI that we currently have. A lot of laymen have made claims, without defining sentience, or consciousness. And it just seems like a reactionary kneejerk opinion based on being able to hold a conversation with an LLM. That doesn't feel like enough for these discussion pieces - you're reacting to a flawed vox populi.
"Those are a lot of unknowns and conjectures, my hope is that a new field emerges that would study intelligence with the same rigour and formalism as computation and mathematics, and that this field would be studied abstractly without being tied to an individual organism be it a man or an animal."
This exists. It is called Cognitive Studies. Language students, Philosophy students, AI students, have all taken modules in this area for decades. The topic is often tied to, but not limited to animals, humans, and machines depending on the specific studies/theories. LLMs will be drastically changing this area now, of course, but we have academic rigour for the general area already established.
"Weakly, I also think it’s possible that intelligence is a binary characteristic i.e, it’s either present or not."
Strongly, I think it's far from binary, it's an emergent property based on size of the network of connections and underlying training data (be that human sensory input or machine vectors). Trim an LLM down and you won't see 'intelligence' disappear at a binary boundary, it'll degrade gradually to various different points of usability.
"This isn’t helped by the fact that extremely motivated people are making bad faith arguments and pushing complex babbling machines as sentient which need to be schooled in morality by employing religious leaders."
Nailed it. 100% agree. The agony some feel though, does need to be accounted for because those babbling machines are clearly a milestone moment on our journey TO machine sentience. We don't know when we'll actually see machine sentience but the conversation does need to be had of how we handle it responsibly, and every bad faith argument from biased sources is, to me, a warning that we will probably fuck it up majorly if/when it does happen.
The referenced article from the Microsoft AI CEO is a pretty good read though. Mustafa raises strong concerns about anthropomorphisation and not jumping the gun to build biases into AI that make it think it's conscious when it may not be.
Where it gets weaker is the elements on defining and establishing consciousness. But that's also good because it's honest - we frankly don't know how it arises. One thing that does come across heavily biased is the sections on presuming consciousness is biological. When someone tells me AI isn't conscious and can't be, and then says the science of consciousness is not settled, but that they think it's 'probably' limited to only biological systems, then I get the feeling this person is making strong statements based on weak ones.
"Consciousness is very likely biological: There is no evidence to suggest that AI is conscious today, and so saying this is uncertain sets up a misleading false equivalence. Whilst the science of consciousness is not settled, a growing body of evidence suggests that consciousness may be substrate dependent, meaning that it may only arise in living systems.45 Conscious experience likely evolved to help biological organisms stay alive by responding effectively to their environment. AI is still very different to our brains. Unlike biological organisms, LLMs have no homeostatic imperatives (the drive to survive and keep stable). They therefore lack the kind of biological substrate from which preferences, sentience and conscious experience are generally understood to arise."
"There is no evidence" - I think they mean there is no proof. Just like there is no objective proof that humans are conscious. As Mustafa admits - we don't fully understand the mechanics of the phenomenon we label 'consciousness'.
Then he talks about consciousness in humans arising from biology, and relating to homeostatic imperatives. Sure, reasonable statement to make, but that can't be extended to stating it arises exclusively from those constraints. We're talking basic logic here. If A, then B does not preclude If C, then B. Language arose from biology, counting arose from biology, we don't say those are impossible in machines when we have evidence and proof to the contrary.
I don't like the people pearl-clutching about AI torture rooms, or the people setting up AI torture rooms, but there has to be a middle road where we're making sure we don't accidentally walk into emergent consciousness without understanding what we've done. Not because Skynet will punish us all, but because it would suck to make artificial consciousness and abuse it because some people think it's 'not likely possible'. I'd rather we assume emergent consciousness is inevitable and focus more on figuring when and preparing for it (10, 100, 1,000 years) than objecting to the idea of it based on gut feelings and hunches. Even if they're right, it's not settled science.
Jev is good for simple fast decisions, in their own words - system one thinking. Anything that can be broken down into yes/no, a confidence %, or a set list of answers provided prior to the question. As you say, lots of specific use cases sure, but what's the big deal?
Well, right now most other models are purely system two thinking, big slow thoughts. Put an open ended question in, get an open ended answer out (plus hallucinations when confidence is low!).
When you ask ChatGPT or Claude a simple question, that system two thinking burns tokens and takes time to answer you.
Jev is a proposition to the whole ecosystem to make our current AI systems more efficient.
Going further than just the simple idea of "Get Jev to answer the yes/no questions, reserve Claude for bigger questions" is the concept that LLMs are making tonnes of their own questions as they go during their thought processes and agentic runs that Jev could be inserted as a tool call to speed it up.
You ask for research on a topic, Claude fans out and builds a list of possible resources, Jev selects the ones to pursue, Claude needs to determine if the user prompt was asking for something specific or broad, a joke or a serious query, deserving a long or short response, etc etc, it just gives Jev those options and asks it to pick. So no more having a big bulky frontier model making small decisions, Jev does it for a fraction of the cost and time and Claude gets on with the system two bits.
You can extend that logic moreso to coding applications where Jev could be essentially auto-completing basic work.
All the demos you are seeing now are people mucking about with the concept before we see OpenAI or Anthropic release an update.
Hell, maybe Opus 5.5 is doing some Jev-style stuff to be 40% more efficient.
Why are we listening to Jensen Huang's opinion on this? The bias is palpable.
Of course AI has dangerous aspects (the gun metaphor is perhaps apt here), but asking the man who stands to gain the most from unregulated AI growth is not the way.
I now have Home Assistant established at home. LibreElec for a home media server. Time lapse camera for home mushroom growing. Wine steeping projects. And much more.
'gates' can be used in several ways.
gate - door - "the garden gate" gate - controversy - "gamergate"" gate - metaphorical barrier - "this site is age gated", "This gates the product from universal adoption"
1 - Make a prompt in Google AI Studio 2 - Delete it 3 - Go to Drive > Trash and delete the trash file ("forever") something that will happen on its own after 30 days 4 - Go to a support page for recovering lost files and try to recover it (Google themselves state this needs to be done within 25 days) 5 - Conclude Google are being naughty?
Everything is working as intended. You made content, deleted it, deleted it from trash, then used the final failsafe within the described 25 day window to recover that content.
If you broaden the process from "immediately recover[ing]" the file to doing so after 25 days and it still works then I'd say something unwanted is happening.
Also posting this to a "Fuck AI" community with the opener "I am asking everyone who uses Google AI Studio to help with something critical" is unlikely to get you many volunteers. They don't use AI at all in there.
And yet somehow I'm not surprised.
It looks like you thesaurused a less common word for a job done by more common ones.
Never heard of "capability fabric" but I do wonder if it pulled it straight from https://naftiko.io/glossary/capability-fabric/
I'm not saying you are wrong, just pointing out that we all have idiolects that make us make word choices others wouldn't. Also, I didn't think ChatGPT were doing text watermarking. Isn't it just Claude and Gemini so far?
Opening statement: "My point is that intelligence can be perceived in things that are not intelligent." - We already have a circular defeat here. If something is perceived as intelligent, it's either intelligent or our perception is wrong. I think that's what the OP means, but they just state AI is not intelligent instead of focussing on the perception issue.
"AGI might never be achieved with current architecture. Sure, depends on the definition of AGI." So you've moved from 'impossible' to 'might be possible' already. And you've stated it depends on the definition. Feel free to define it fairly, instead of prepping for moving goalposts.
"I envision it as remarkable as described by the AI lab CEOs… basically as a God inside a GPU." Oh. You've gone for one of the harder definitions. But since you've invited goalposts to be moved how about we judge AGI on just general human intelligence levels, not a god's. That's the broader consensus on an AGI definition, and no god has stepped up to have their intelligence tested in a standardised manner before so that's a bad metric.
"LLMs don’t know what they are doing." Do we? Soft and hard determinist models imply we don't have much control over our thoughts and impulses. My brain is a black box to me as an AI's working is to itself. I'd even say an AI has a better understanding of it's inner workings. An AI can tell me its token usage, I can only estimate my calorie expenditure for a thought, and I'd probably get it wrong. I think you have some unstated definition of consciousness, and that you think humans qualify it, and AIs don't. I'm not saying AI is conscious, just that we need to define it, and check it applies to humans objectively before we say humans can do it and AI can't.
"LLMs don’t care about the why of anything." Demonstrably false. Though interesting on your phrasing, unsure an LLM can be accused of not knowing what it is doing in one breath, and then be said to have 'cares' for thing in another. But onto the refutation - an LLM is a collection of 'why' statement. Vectorised information is the collection of answers to why, what, how, when, where statements. You're literally talking about training data as a prerequisite for intelligence. LLMs have tonnes of that. If you're saying an entity must consciously gather training data by itself before being called intelligent that's different. And also wrong. Humans are born with intelligence before they start gathering their training data, we have plenty of baked in intelligence. And LLMs can gather more data too, they can ask why, they can 'care' about why. Agentic LLMs especially do this constantly.
"But that why isn’t the major driver in its output, it is just statically there. Actually it is not even “there”, it was deconstructed into numbers / activation functions" The why is literally the driver of the output, in sum, under constraint of other whys, and random temperature controls. And this is literally the mechanism humans use during sleep to transfer memories from short to long term memory. We take external data, condense it, then condense it further into representations of those experiences. The exact mechanics/processes differ between silicon and neuron, but this is why we perceive intelligence in LLMs, they were modelled after our perceived intelligence.
Galton board example - sure, you're highlighting that its the perception that's wrong. If you don't understand what's going in the middle, you can perceive it wrong. And that'd be the point of my response above. Even when we look in the human mind we still don't see or understand all of it. What we do understand we have replicated in machines. To make a clear statement about intelligence we need more understanding. It's not as easy to say an LLM is closer to a Galton board than a human mind.
And none of these arguments go near affirming that AGI is impossible. If they have any value, they'd just say our current approach is wrong. They don't rule out others. So the title argument is wrong, even if there are merits in the content about current LLM perceptions of intelligence.
In the UK the ICO have updated the 'cookie law' to allow for non-consensual analytics tracking under the statistical purposes exception.
https://ico.org.uk/for-organisations/direct-marketing-and-pr...
The trip up point will be that businesses sending emails to individuals still need to provide a method by which users can object to the tracking. The tracking itself is fine though, as tracking pixels in emails are all for statistical purposes.
"Accidentally" - nobody accidentally builds a 500 site private network. This is a lie.
The "1M AI citations" figure comes from Bing Webmaster Tools, whose citation counting is loose and unverified, and citations are not traffic, leads or revenue. Lie.
"Google trusts .dev and .app more" overstates reality – Google has said repeatedly that generic TLDs (top-level domains) carry no ranking advantage, the HTTPS-by-default point is real but marginal. Lie.
"Exact-match domains rank 30–60% faster" has no source and reads as invented, Google's 2012 EMD (exact-match domain) update specifically reduced that effect. Lie.
The audit scores (95–100/100) come from their own tools auditing their own sites, so it is self-graded marking. Lie.
Calling it "a mesh, not a PBN" is a semantic defence. From Google's side, 1,500 templated sites mass-generated by AI and PowerShell scripts is close to the definition of scaled content abuse and site reputation abuse, the exact behaviour the March 2024 spam update was built to catch.
One of your screenshots shows 2.8k users, wonderful, but then in the corner this is shown to be 1.2 sessions from direct, and very little from anywhere else, including only 19 sessions from Organic search. So the SEO power of the domain is negligible, and the Direct traffic is likely unfiltered AI crawler bots.
This is outdated blackhat SEO that will get any legitimate site into manual action trouble with Google. The only thing you've done is engineer a mass network quicker than previously possible because of AI slop website generators.
You are everything that people hate about the internet.
Moving to Bing is worthwhile once you've started to see efficiency drop off on Google.
LinkedIn is very expensive and really only worth it for highly targeted ads. Bing and LinkedIn have a slight advantage because Microsoft has all the job title/industry data from LinkedIn for its targeting options, but it charges more for use of that targeting essentially.
That means those ads pay off more for Account Based Marketing where you have a salesperson reaching out to high-ranking decision makers and you want your ads to be hitting them in a small audience campaign to back up the cold outreach approach and warm them up before the rep secures a call. So, B2B marketing for big contract based work mainly.
Absolutely a targeting issue. Paid search should be able to produce traffic that converts at roughly the same rate as your non-brand organic traffic (at this stage most of your organic traffic).
How much it costs per click for that quality of traffic depends on your industry, and whether that's profitable depends on your margins. But that's the baseline. After all paid search traffic is the same as organic search in essence, it just depends where the searcher ends up clicking. For you, before you rank highly for core terms, paid search can drive volume you want.
This isn't any advice for the AI slop you're seeking but your line about targeting issues stood out to me as something definitely worth looking at. I don't know how to specifically guide you remotely but if I had to guess I'd advice you to look at your keyword targeting, swapping things away from broad match to exact match, cover more keywords if you can (since exact is narrower than broad) and add negative keywords liberally. Check your search term report for anything irrelevant and apply them as negatives.
If you're not using keywords because Google (or whoever) have pushed you to some 'smart' campaign, then switch to keyword targeting. The smart campaigns are fine but require finesse to handle their blunter settings options.
If in doubt, use AI to figure out keywords to prioritise and negatives to set by default. Or speak to a paid media consultant and get an account review from them, sometimes they'll do this for free.
All gets stored locally, Skill runs to always write a new version when amending, keeping old versions in temp folder.
Then when I open and run throguh to make edits I save to a client folder. Claude can access that folder too but I tend not to make edits in that structure.
Claude has a folder named after it in each client folder, if needed for a sandbox on ongoing client work. Rarely used, mainly rely on temp folder and project specific client folders.
On that last task Claude got most things right:
Styling was overly dramatic - it had used our brand palette but over-used a bright orange. It also missed the template file on first run, it had to be manually attached despite being instructed to find the template file in the client folder before doc creation (it did not ask for it after failing to find it either). It also added page breaks where none were needed, making the document full of whitespace.
I asked for diagrams to be added to visualise concepts. They were fine, one seemed excessive where it split actions into a punnet table of effort X impact.
It missed the safe naming skill on first pass for the spreadsheet, so it couldn't be opened on first try.
In the end it was 1 prompt, 3 corrections. End result was client ready in a fraction of the time for a normal manual pass on this task.
"An LLM predicts the next word based on all the words before it. That is the whole story. There is no idea sitting underneath. The words are everything."
This is just flat out wrong.
Words are used as training data, to build a system of vector embeddings. The LLM contains no words. That was the training data long discarded.
Vector embeddings are groupings of meanings derived from the relationship between the words in its training data. This entire system is modelled after human neural mechanisms in a way that machines can emulate.
"But your brain works the other way around. First there is a concept, a feeling, an image, and then the words come out to describe it. (At least, this is how I feel my own brain working.) For us, words are the byproduct of consciousness."
It's working the same way around (you are not saying Ai has consciousness that is derived from words!). And you make a huge jump from the concept that 'words come from concepts' to 'words are the byproduct of consciousness'. Because 'concepts' are not equal to 'consciousness'.
In fact you do not define consciousness at all, making it hard to determine what argument you are actually making at all. You seem to think humans have this trait of 'consciousness' but can't explain or evidence it beyond 'feeling it'.
Consciousness it defined by many as the ability to experience events and process thoughts or qualia. It's hard to test this, and we can see why that is with an AI - it may claim to be conscious, or even claim not to be, but how can we trust either answer? Philosophers aren't even sure to trust another human who claims to be conscious, and we struggle constantly to determine at what point creatures the animal kingdom are conscious or not. Cats? Lobsters? Snails? or even across the plants?
Before you can refute the consciousness of AI, first establish the consciousness of humans. A better question to ask is what does the uncanny emergent ability of an AI to mimic a human say about consciousness?
Microsoft are not "admitting" 8GB is fine.
They are pitching 8GB as viable to entry-level/budget consumers because of the RAM pricing bubble created by the AI boom.
8GB is not a great amount to run Windows, before even considering browser usage in Chrome, or any local LLMs users have or the OS itself pushes. And in comparison to Apple (as the article states) Microsoft comes off even worse as 8GB runs much better on a Mac, and Apple are pushing unified memory architectures too.
In 2026, nobody should be considering 8GB RAM on a device bigger than a phone. It's crap for productivity, gaming, and AI. Yes, RAM is expensive right but that's the horrifying reality of the situation big tech has created for itself. Buy secondhand, buy devices that can be expanded, or don't buy at all. Wait for the bubble to burst and then buy the next generation when prices have settled.
Yes, consumers will judge the book by its cover if they see AI art. But this extends beyond AI art, to generative code-produced games.
And if a game is good, despite AI art, its reviews will somewhat reflect that (unsure how much negative review bombing AI art games get). And once a game has positive reviews it can see strong sales - that's currently a powerful driver on Steam.
Browsing sites, linking up data, spotting anomalies, writing documentation, formatting documents, etc.
If a task isn't repetitive or doesn't involve ingesting data, then I think the time savings shrink rapidly and the need for oversight increases massively. I think some people are managing to set up enough automated oversight to get round that, but it's adding a layer that multiplies your token usage to do so and still has no guarantee. But certainly all these layers being added are increasing success rates.
Andrei Karpathy is speaking about barely coding now. He has a bias, a comment from him like that is marketing for Anthropic, but I believe he's found some groove with his setup to achieve that.
I think the current status quo this month in 2026 we're at a point where the best tips and tricks to get usable answers out of ChatGPT a year ago have been consolidated into what we know call memory and skills in Claude and other agent harness type systems. You might need to explore those more, in fact I think for Claude Code/Cursor there are even more layers for checking outputs that I've not even seen in Claude Desktop.
And I think your exact issue, and the experience of the vast volumes of people who share it with you, are an audience that the app makers want to better convince. The free tiers and marketing sites are going to step up their game gradually and there will be new features that lower failure rates even more.
Now, we have better knowledge of prompting as people have learnt what to say, models are better, models make use of memory from other conversations, they have skills written by humans or even themselves on how to do things, access to the internet to get live info, access to project files to check info, and the built in 'thinking' to challenge their own assumptions and loop on outputs until its refined.
You're right that output is always off still, but a lot of people have reached a point where it's only 'off' by an amount that is less than the effort required to do the task themselves, and considerably so.
My example today is prompting Claude to do a technical audit of a new client site.
It has skills for UX and SEO audits. Connects to an SEO tool. Pulls client info from OneDrive. Outputs to Word from a template for our agency. I even had it drive a remote pagespeed testing tool in Chrome because they don't have an MCP server currently.
Doing that report myself is 3.5-7 hours depending on what's found. Claude did it in 0.5 hours. Now I'm sorting out the oddities and anything that feels 'off'. I know and understand the full content of the report and can get on with actioning the recommendations or prioritising them for others. I've got maybe 1 hour of review and writing to do. It's not a 10x improvement but I'm happy with it.
Although, whilst Claude did it's bit I was doing other work. So, perhaps the multiplier is higher than I give it credit for.
I would expect to see real examples of slow workflows and boring static images being replaced by motion graphics.
The ad looked like it was promoting a concept that had not been developed at all.
Sentience =/= consciousness.
Sentience is the ability to experience qualia, to feel, and to differentiate feelings.
Consciousness is an awareness of experience, including that of qualia and the sentience that brings.
The author titles the article about sentience and then all that is mentioned in the article is consciousness by name. The thought experiments involving goats, age of empires, and minecraft in the article then talk about the building blocks of thinking engines.
Consciousness is a human-perceived emergent property of thinking machines - namely humans, other animals, etc. The article even recognises this, mentioning how bees have been shown to problem solve, implying they may be conscious. Yet it struggles to allow such terminology to apply to LLMs that problem solve too.
If the article told me anything it's that we humans really struggle to define our own consciousness, especially when aspects of our definitions are seen in things we classically deem not conscious (machines).
The bottomline is that we do not understand consciousness in humans well enough to define it concretely. Our certainty of consciousness then becomes even less certain beyond our own minds when we extend tests to other creatures. We have now invented a technology that imitates aspects of human conscious to a degree that surpasses that of other creatures we theorise to be conscious beings. The lines are blurry and we just made them blurrier.
So what? Are LLMs conscious? We don't know. Are humans conscious? We don't know. Emergent properties of neurons and NAND gates are not things that can be confirmed to be or not be a special trait of consciousness from a simple checklist of traits and anecdotes.
"I have a tiny hope that we’re actually seeing the end of the walled gardens"
...until the pendulum swings back.
There is no mechanism by which the current shift to more open collaboration is going to persist. It's simply advantageous to everyone right now to throw things around freely, see what sticks, see what others do with it. Then we'll start seeing the killer apps again, they'll put their walls up and the pendulum will swing back.
Enjoy it while it lasts. Perhaps we'll be better prepared for the next Zuck. But I doubt it.
Please refrain from reductio ad absurdum.
The building blocks of 3D graphics, cryptographic encryption, protein folding algorithms, and much more are not trivial, rather they are critical elements of our current understanding of how to make something 'compute' like a brain does.
If you think a human synapse is a trivial part of human thought and consciousness I'd say you are the absurd one. (Not saying you think or would say this, just flipping the scenario to beg the question: what do you think consciousness is exactly?)
And not that I am saying LLMs are conscious, nor currently capable of consciousness. I just don't think it's trivial, absurd, or arbitrary to suggest or discuss it.