Is anyone actually using Gemini directly?
It sucks total bum compared to even some quality mid tier models on huggingface. Even the joke about ChatGPT was that it performs better than Google’s equally bum SEO’d search engine.
I feel like Google is suffering much less at throwing it into all of their apps than MSFT doing the same with copilot.
It’s still a leading long context model, IMO.
But I think it feels dumb because the default “harness” is pretty simple and dumb, more than the base LLM.
That, and it’s really awful at 1 temperature (which is its default).
Maybe some leadership is guzzling Kool-aid over how bad Google search has gotten? Because it leans on that pretty heavily, without many other tools.
Mistral or Lumo way to go
Yeah, Mistral is just… not very good anymore.
Even strictly compared to open weights models. Everything I’ve tried feels obsolete.
I don’t know what models power Lumo, but Mistral is so far behind the competition it’s not even funny.
Mistral Medium 3.5 isn’t that far behind comparable current open weight models.
Maybe, but Mistral is a commercial company that offers commercial products that can’t really compete with OpenAI, Anthropic, and the others. That’s what I meant.
I’m curious as to why you say that? I use Le Chat and to me it feels exactly like ChatGPT or Claude, it can code well, translate, search online, everything.
Personally I really struggle with it. I wanna use it because it’s not from a US company, but it’s just wrong in like 60% of whatever I ask it. Sometimes even when corrected. It’s good though, makes me stop using the clanker and actually search for myself. Sometimes I do resort to ChatGPT or accidentally invoke Gemini when searching though, and those two have a way higher hit rate
My two cents on the matter is that LLMs are incredibly useful, but I would never use it to that end. I never ask about facts or use it as a search engine. Sure, use it as a springboard if you don’t know where to start or what to search for, or maybe to find the right link to a website. But directly relying on it for information is risky, whatever the model. Maybe this is the reason our experiences seem to differ.
Subjectively, it feels similar to models we used a year or two ago. Not that drastically different from what Anthropic and OpenAI offer today, but slightly worse. For instance, for complex coding tasks it offers basic solutions, while Claude often offers more options and details - as if it knows more.
Objectively, benchmarks. Mistral looks comparable to other open weight models (as another user mentioned), but not as good otherwise.
With the results of chat out of the box I kind of agree with you that mistral feels behind.
However, there are some features that I really like and make the experience even better than chatgpt.
- Agents are pretty cool and with some setup produce very good results
- managing libraries for documents/context is better than in ChatGPT. Also adding specific libraries to agents is nice.
- Scheduling tasks has just been added and I want to try that out.
(I have never tried the paid version of any LLM chat so I can only compare free tiers)
I see, that’s about the time they all got so good that I stopped trying to keep up with the latest benchmarks. It works perfectly for my needs so I definitely wouldn’t dismiss it for anyone wanting to switch to a European alternative.
Sure, totally depends on your needs. But it’d be great if we had one of them frontier models in Europe.
True! I’m loving lumo so far
Who the fuck cares
I care, thanks for asking.
That depends; it will most likely cause some pretty serious issues outside the AI space. When that bubble finally pops, it’ll probably make the dot-com bust look like a bathtub fart. Knock-on effects over non AI industries, and a lot of little bitch CEO’s that put all their money/companies money in AI will now be out a whole lot of cash and go on (more) firing sprees.
And if it doesn’t pop, it has made ordinary people into millionaires. I was having a look at Swedish Avanza site where they launched a new feature where people can share their portfolios (what stocks they are invested in, and their account growth). The best investors got in early on all this stuff. Nvidia and now memory and disk stocks, so they consistently made tons and tons of money.
All while people has been waiting for the bubble to pop since… 2021? Maybe earlier.
We absolutely don’t know if it will pop, and even if it does, there was a huge opportunity to make money from it over years and years.
Thats what investment bankers did, and that’s what ordinary people did. Some got very rich on this.
Im just saying that because it’s important to realize that the way we think may not be optimal for our lives sometimes. I understand fear. But I also understand that some risks are necessary to gain something valuable.
Chat GPT came out in 2023. To the extent there was a bubble in 2021 (crypto and EV come to mind) those frothy parts of the market have deflated.
Thought it was worth adding to your comment that getting rich by taking extraordinary risk may not be “optimal”, and that looking at upside outliers is a bad way to decide what to do with personal finances.
If what you’re advocating is buying well diversified, low fee, investments with a long time horizon then carry right on though.
You can adjust your risk, sure. Those low fee diversified investments will give you index level gains, sure. But I meant getting actual rich. Then you need to bet on the right stocks.
Me. I care.
Two certain clowns, though for different reasons
It’s a weird business where everyone is offering an environmentally unsupportable service at below cost, hoping to outlive the competition.
Market share of a negative profit market.
Market share of a negative profit market.
Yes. That was my reaction, too.
If I start giving away autographed headhsots, have I then also cornered an emerging market? lol.
It worked for netflix and the steaming services. Now terrestrial (cable) is dead and adsupported streaming tiers have returned lol.
It’s how every tech company that “disrupts” a market or indistry works. Uber started of burning shit tons of cash operating at a loss till it replaced enough Taxi services then jacked up the prices.
The problem with AI is that they cannot increase the prices enough to be profitable. The AI companies are waiting for future hardware tech that will be energy efficient enough to make AI profitable before they run out of capital to burn.
The problem with AI is that they cannot increase the prices enough to be profitable.
I saw something about the SpaceX IPO that said for it to be justified at that price, everyone on earth with some sort of money (they defined it as earning at least $14,000/year) had to become an xAI consumer and spend $28,000/year. Seems reasonable /s
Wouldn’t that hardware then be priced proportionally?
I was thinking about that, but it did make governments move to digital programming (eg BBC, NPO), which they likely wouldn’t have done if it wasn’t for Netflix. I think it also improved the internet speeds here in NL significantly.
That’s not correct - the BBC announced iPlayer in 2003, tested 2004 onwards and finally launched in 2007 after being delayed by lobbying. The iPlayer was held back from full launch due to concerns from commercial competitors - in particular broadband providers lobbied against the iPlayer service because they feared the “pressure” it would put on the broadband infrastructure.
Netflix launched their streaming service in 2007.
Netflix did not originate the idea of streaming (nor did the BBC to be clear), much like Apple didn’t originate the smart phone. Netfiix did however do it better than it’s competitors, particularly the incumbents in the commercial sector.
I remember iPlayer in 2007: it was a mess, using DRM and P2P, requiring a full download before watching it. I wouldn’t call that a streaming service at all. Shows had to be watched within a week of broadcasting.
I welcome feedback but starting a comment with ‘thats not correct’ and then blatantly being incorrect is just some ol’ bullshit.
I never mentioned that either of the companies originated the idea of streaming. I only posited that Netflix pushed governments (in this case the Dutch and UK) to move to digital programming.
Painful memories of drm do come haunting back, but I do wonder if any streaming service was much better at the time?
They did a better job by offering an unsustainable variety of programming from all the studios in one place at the same time. All of the competitors at the time only offered financially viable services.
I believe in ai now because I looked at the Netflix balance sheet and thought. “There is NO WAY they could become profitable they are spending wayyy to much money and have way to much debt. It’s financially impossible to get out of this hole”
I understand how it worked and how it could not have. There are a lot of ways this could fail on AI but there are some real ways forward. AI has a similar application reach as the internet. It’s world changing.
I see why meta and google are going in hard. They lived though the rise and fall of blockbuster. They saw Sony release 3 different steaming services before after and during Netflix. This is the disruption for the current generation of tech and their revenue model.
Someone is going to ‘Win’ AI and a lot of others will loose.
One company may win, but we all will lose for sure. Specially investors, the math doesn’t add up even if revenue breaks even.
It will be even more funny if these companies are declared financially bankrupt and the profitable parts are sold separately and the non profitable will stay with share holders.
I have never seen that happen. I have only seen companies get bailed out. It’s been decades since a giant insolvent company was split apart and sold for pennies on the dollar. They just enough money to buy financially solvent competition
Wait until the winner shows up and proceeds to absolutely wreck everything with fees and subscriptions jammed into everything remaining that didn’t need AI.
Look up commoditizing the compliment.
The free market strikes again.
So, people are falling for Anthropic’s marketing scheme?
The company taking most of chatGPT’s market share is actually Gemini, which I think is cheating because it’s basically just padding the numbers with random google searches.
What scheme, their last models have been so much better than OpenAI’s it’s no surprise people are moving to Claude.
Personally, I find myself using the Chinese alternatives more and more as they are just way cheaper.
The good thing is that a deepseek can be run locally relatively well with consumer hardware. I trust chinese companies as much as i trust american companies with my data and my prompts.
You have 170+ GB VRAM at home? (:
I mainly use DeepSeek v4 Flash now, it’s the cheapest around and the quality is high enough for coding. At work we’re throwing tons of money at Claude, but even there I usually stick to Sonnet (as Opus is burning money).
You don’t need 170+ GB of VRAM. Whole model can be run at around 1 token/second on a modern hardware from an ssd. Which is slow, don’t get me wrong, but it still somewhat useable.
Upd.Once again, for those who use AI because struggles to read: it is slow, but it is usable. Which is, by definition, means that you don’t need 170+ GB of VRAM to run this model. Period. It runs from ssd. That is a fact.
1 token/second isn’t remotely acceptsble lol
“Somewhat” is doing a lot of heavy lifting there 😂 How much time does it take to process your average request?
I’ve been loving minimax-m3 since its release
Theres a lot of good models free online that are open source/Chinese and the only cost is a bit of slowness on my rig. No token limit or whatever. Totally agree.
Its about as good as commercial products now…
Can’t be Europeans as they’re locked out from the new models LOL
As is everyone?
I switched to opencode and deepseek
AI is becoming a commodity. Companies with such high burn rates will have a hard time staying alive.
Especially with open weights models available from cloud AI providers.
Yeah the open weight models are getting good enough that a lot of people can just host their own for most tasks.
On what hardware? I have a 1yo laptop running Linux and it’s slow AF on open source llms. Might be to do with video card drivers, but I’d be happy to hear suggestions on which machine to put next to my raspberry pi for a local LLM.
You don’t really want to be running them on a laptop. You either need a big dedicate graphics card (with 16 GB of RAM preferably) or a unified RAM/VRAM machine like a mac mini or a strix halo.
Local model running is definitely more doable now that it was a year ago but its not yet at the level of throw it on anything and it’ll work fine.
I’m talking more enterprises running specialized hardware, but if you needed to do it on a laptop, your best option would probably be strix-halo.
not a gambling man but if i was… https://polymarket.com/event/sam-altman-in-jail-by
This is my first time actually navigating to one of these websites. God damn it’s so much more dystopian than I even imagined.
For those who didn’t read past the headline: OpenAI is still growing, but competition is growing faster.
I wish headlines would start to qualify market share. They had over 50% of the active usera/subscribers/acticity or something like that. The addressable market on the other hand isna lot more people and they probably never got anywhere close to 50% of that.
So ChatGPT, DeepSeek and Claude, the rest is residual. What a shame. Mistral Vibe deserves better recognition.
I only used the free version of Mistral and it’s not as great as the free version of Claude, which is a shame because I want to support European AI but Mistral seems to be the only European alternative at the moment.
So OpenAI is becoming more profitable?
This isn’t good news. This just means now the market is fractured with many AIs instead of just one. If anything it’s just a sign that more people are using AI.
Call me when the headline is “ChatGPT shuts down”.
deleted by creator
It is
Less customers same expenses
More chances of them going bankrupt
If they do it’s probably going to be Musk that buys the corpse.
This just means now the market is fractured with many AIs instead of just one.
Fundamentally this feels like it should be a good thing…
But it does feel like this just accelerates things.
It’s great news. The only hope for these companies is getting to monopoly enshittification before investors give up. This shows that timeline might be impossibly long.
If OpenAI’s numbers are true, more usage means more cash burn. And more competition means companies have to fight to offer the best deals to customers, meaning more cash burn.
Now if the market was reasonable…
Now if the market was reasonable…
lol. I am the market, and I am reasonable.
I’ve got actual use cases for AI, and I’ve got a local model running on my own hardware.
In short - I’m the long term buyer after the hype wave ends, and I ain’t buying shit from these assholes.
That said, if a nice fully open source home appliance gets announced, by a company that hasn’t burned all available goodwill and trust with random acts of assholery, I’ll probably join a wait list for it.
In other words, I’m saying there’s a chance.
Unfortunately, unless you’re a business yourself, I don’t think you’re the market anybody castes about any more. The entire RAM crisis was caused because individual people got deprioritized compared to the promise of some revenue from big businesses…
It’s a weird club of CEOs telling each-other to buy their stuff in return for investment.
I’m not sure there is an actual market, once the money-seeking-profits spin cycle ends.
The market shouldn’t even exist.
Imagine if I offered a service to connect people with local serial killers. Then after some time more people started offering that service.
Would you say “Well hopefully competition will drive down prices at least”?
Yeah big AI needs to not be a thing. Small, focused, FOS, and local AI seems like a future with an okay compromise. I say as a very anti-AI person, but for those that use AI, that seems to be what they’re doing, in my vicinity.
It all still burns energy at a time when we should be reducing our energy consumption. And for mainly frivolous things which we could have done without AI aynway.
It does, but less energy, and it distributes it worldwide rather than destroy the ecosystems where their data centers are located.
