First, a web/cloud version of my fav, Qwen. I use it most to make bullet notes & observations of text, make light edits on my writing, and study humanities & literature.
Some gimmicky/interesting ones:
- claudelives: an old version of claude brought back by fans.
- berduck: hepful rubba duck friend 🤔
- Kobold lite, an RP bot run by volunteers who take turns. it’s all a local LLM on some nerd’s computer.
- LM arena: a head to head comparison of the same prompt on different big cloud services.
Try something different.
Deepseek: To me, the reigning champ in big LLMs on the cloud. It’s quite detailed and accurate, as far as I can tell. It’s also notably less complimentary/sycophantic than Gemma (and also Gemini, I assume) and Llama.
From my POV, Deepseek hallucinations are kinda predictable; It makes up answers when it really doesn’t have the info to know. but it doesn’t as much impulsively make shit up (It still does that sometimes, though, so do actually check the facts you get from an LLM)
If I try to use built-in translation tools on social media websites, they will probably tell me to use Deepseek for translations, instead.
Perplexity: I’m just starting to use this one, checking it against my searches with Google Scholar and other attempts to limit searches to reputable sources. It uses a lot of weird random blogs as sources, so I often tell it to only give me results from universities or publicly owned research institutions, or define for it some other segments to backup it’s arguments, like certain schools of thought in philosophy, etc. It helped me find some papers I knew I remembered existed but couldn’t find anymore.
Le Chat: Simple (this is a good thing–also more optimized, less hallucinations) and fast, and hallucinations kinda (to me) come off as opinions which is a bit charming.
This is a great tool to reduce the complexity of the text you paste into it. providing summaries and returns in a relatively natural writing style (for a machine). If other LLMs are too wordy and redundant, try this one. And uh, if you like to be complimented by the LLM, this probably ain’t a bad pick either, but I think you should also ask for prozac or something.
Quick edit on the environmental impact, which is widely misunderstood. This supposedly environmentally conscious part of the AI debate is largely nonsense. An individual accessing the service does not add significantly to the energy, water, etc, consumption of data centers. The models consume all that energy during training, and are relatively cheap to keep using afterward, which is why I can even run small ones on my own hardware.
Past the point it takes to make 14b Qwen 3 in the first place, me downloading it and running it requires no more power than downloading and running Cyberpunk (potentially less, depending on my prompts). making the LLM is the hungry part. just like everything that is made or consumed. but liberals are easily manipulated by literally any sufficiently convincing napkin math and become this:
so, imagine the LLM is a cake, baking it takes 99% of the energy to create a cake experience. Eating it, or no, does not change the energy required when it was made. Worrying about the cost of the cake after it comes out of the oven is foolish.
If these LLMs weren’t private companies competing against each other, there’d be much less need to train up a bunch of them, that are only slightly different than the others. This is the innovation neoliberals asked for. literally all of industry is doing the same wasteful, cancerous growth trajectory as the tech companies. which is why we are all headed toward multipocalypse.
So the environmental angle is a flimsy, dirty trick. Organize boycotts and unions, and don’t buy things when they are part of a boycott. Act in a disciplined, not erratic way. Because no one is adding up your woke points from your purchases or lack thereof to pat you on the back.
Loading comments...