Відео

Expensive? I host things on digital ocean. 5$ per monthn for cheapest droplet. But i host bolt on cloudflare pages for free for now. One of community members shared how here thinktank.ottomator.ai/t/deploying-bolt-diy-with-cloudflare-pages-the-easy-way/2403

@EduardsRuzga Thank you a lot! Much needed this ❤️❤️‍🔥

@@newbie_productions its probably due to context size What is project you are importing, what llm are you using?

@@battleabhishek what model are you using, can you export snd share the chat?

Sure give me ur email id

That is exciting thing about this. Now non developers can small things onky developers could before. I trough my life often did small projects that automated things in my life. Helped me a lot. Now more people will be able to do that.

Most welcome 😊

@@sqyer thanks. I do want Bolt.diy to get here too. Be able to run tests. Sadly we cant run cypress, playwright or Puppeteer. But there are other options.

Yeah, sorry for that, its know bug, I will try to get to fixing it tomorrow. I mean that '''jsx''' part As for second part, that one I did not have yet... Can you export the chat and share on github gist or something?

@@EduardsRuzga when i said it to edit the code it said "im just a chat bot and i can't directly access your code files "

@@gamesofaslandev I see, try asking it "use bolt action syntax" to edit files We need to update prompt to fix this

@@EduardsRuzga Worked i asked it to remove the ``` stugg using bolt action syntax ty

@@EduardsRuzga btw it would be really handy if i would be able to know all the syntax :)

Hey there, since this video I switched to variant where you need to get your own keys. I recommend google ones. Link is here bolt.myaibuilt.app/ It supports latest google models like gemini pro 1206 and gemini 2.0 flash They are free with big quota and quite good.

Sadly I am not aware of anyone providing that model for free. Here is list of free providers, it may be incomplete. github.com/cheahjs/free-llm-api-resources Best models for free are google one, and there is new provider we did not yet add, github, it has gpt4o and some other models provided. Sonnet is not available in any free offering.

@@EduardsRuzga thank you for answer sir okay

@@eta2001carl why xAI and not google or chinise?

@@EduardsRuzga in largest model that is, xAI as x3 the compute and it is in training (gork 3) in execution that is a different story :)

@@societyofminds cope? With what lol? I am playing with them since and they still work very finicky. This is beta release for technical people. One example is Puppeteer mcp It does not always load and blocks orher mcps to load. Same happened with my terminal access mcp and turned out that they use stdout as communication channel. So if anything in your dependencies logs something mcps break... All in all its tricky to develop for. CustomGPTs instead integrate with REST APIs and it just works...

What provider are you using? Google or OpenRouter? If OpenRouter I heard from many people that their free models fail often. I think they have high demand on free models and run out of free quota often... Try using google instead for google models. That is one I am using now.

@@EduardsRuzga openrouter. when go goole i cant find that free version

How so? VS Code on windows helps to install somehow?

@EduardsRuzga yes if you would have opened your project in vs code after installing node js you could have just opened terminal on vs code and use npm install and it would have installed all dependentcies for that project!

​@@madjoker5891but how is that different? Istead of downloading node and project and installing you download vs code, node and installing. Seems like additional step.

@@madjoker5891 that is the same I showed just without VS Code :D After your comment I thought that VS Code can install everything for you in one click. But no... Makes me wonder if we should add some kind of setup scripts :)

@@madjoker5891 But in all honestly, my goal is to make hosted version do everything that local does so people do not have reason to install locally at all

What provider you are using?

@@EduardsRuzga i tried with anthropic -claude3.5 sonnet and openrouter both

In case of Anthropic yes. They do not have free quota. Fully free models are local ones. And some providers have generous free quota for some models. I really like google ones.

Thank you, warms my heart to hear!

Thanks! Wanted to share as I test these thing. Glad its valuable.

Hey Simon. You can find me on linkedin or in ottomator community thinktank.ottomator.ai/u/wonderwhy.er/activity

which model you use after

@@selfdesign7844 you mean language model?

I would say that there is still work to be done to get to level of some of these tools. But we are going there.

What kind of error?

@@EduardsRuzga buongiorno l'ho contattata anche su LinkedIn comunque appena rientro a casa le invio su LinkedIn la foto dell'errore che mi esce credo che sia dovuto al file node perché sembra incompatibile.

Not sure now, but initially it was sonnet 3.5 Its currently best balance of quality, speed and price. But many others are ok too

I prefer Sonnet 3.5 from payed and Google Gemini 2.0 from free

I did one video about that here ua-cam.com/video/-VkGI-cUaWw/v-deo.html Not sure that it will be easy to follow

Hmm, I did 3 iterations with Gemini latest yesterday and it did not got it right for me. Also tried Deepseek R1 Just tried and with prompt "here we have red and green bar, and two shadows, red bar shadow climbs a wall blue wall in the back, solve the problem" + image it did gave me 8. I gave same prompt to o1 with image and... It did get 8 answer too :D Well, my main point is, I know its 8. And depending on a prompt it can get it wrong. How do I know of it was right or not? Is o1 better then cheaper models? I do think that for many workflows its not best tool for the job due to price/performance.

I actually did 4 videos since I speak about reasons for change to API keys here ua-cam.com/video/J5iuC7Te2l4/v-deo.html

Ouh interesting, I want to try that. Was thinking to try some Chinese models on this. Where did you run it?

Actually tried with DeppSeek R1 and its answer was 1.79m :D

@@EduardsRuzga Hugging face

I could have been somewhat ambigious here. Thanks for asking. I base value of pro subscription on ulimited access too all models including o1 and advanced voice. I very rarely run to limits of 4o and o1-preview and advanced voice in chatgpt. 4o is enough and i do not use o1 and advanced voice as much because of no tool use. So if you are not then subscription pro is not for you. As for pro mode of o1. I only referenced AI Explained who did test and found pro mode sometimes being worse than other models. And his trsts mirror my impressions from trsting o1-preview. I hope this clears things up.

@@EduardsRuzga To be honest it doesn't because what is still unclear is if o1 Pro would be able to deal with the "stick with the shadow" question any better.

​@@janklosowski fair, but that was not main point of the video. I will see if anyone bought it for whatever a reason and could test this problem on Pro. Main point was that if you are not hitting limits on advanced voice or o1 as is then Pro subscription is not for you. Pro mode O1 does not change that.

@@EduardsRuzga Ok, but TBH, if you didn't test o1 Pro, how can you say it's not worth it? You show how previous models failed to solve a puzzle. What if Pro solves it better (which is what they promise-that it's better in complex reasoning)? That would make the video's conclusion completely wrong.

​@@janklosowski There are multiple things in this video. 1. o1 is not smarter at solving this issue then 4o or sonnet 3.5, they all fail 2. Why? Because of understanding of images 3. Is o1 Pro better? Probably not because its same model, just used with more tokens per run to think even more. That would not fix its image understanding but I can't claim that. 4. Other part is weather 200$ mode is worth for you. And claim I make there is that if you are not using Advanced Voice and o1-preview already to the limit then its not worth for you. 5. Why I do not use them to the limit? Because they can't use tools. They can't search internet, they can't run code, they can't do anything except for outputing text. While old voice and 4o can and I do use them a lot. Confusion is in you thinking that I claim that 1 2 3 is connected to 4 and 5 I am not saying that Pro subscriptions is not worth it because o1 can't solve this puzzle. I am saying that Pro subscriptions is not wroth it because those models are not yet ready for heavy use without access to tools. 6. I am also saying to be careful with feeding o1 images and thinking its smarter. It gets things wrong. If o1 Pro mode can solve this it does not make 200$ subscription worth it anyways. Tool use for those models would.

Hmm. Did i claim it did not? I said that o1 mini and preview and advanced voice dont. But now new o1 has. Got4o had that for over a year i think.

Oh, got it.

Hmm. Desided not to boost this time. I guess I should have.

@ I’d recommend compression / limiting ** lotta people listen on their phones on the go * headphones or not, and I was trying to listen to your audio through iPhone without headphones ** I could barely hear it even though I could tell the content was good I couldn’t catch much of it

@@josephboomtv7811 i will take a look at this. Thanks.

here are some social network links www.wonderwhy-er.com/ I am also on discord: wonderwhyer Or you can find me in ottomator community thinktank.ottomator.ai/u/wonderwhy.er

Thank you!

I am not in to using docker myself, why do you ask for it if you aren't? But other youtuber did that here ua-cam.com/video/31ivQdydmGg/v-deo.html

Good morning @@EduardsRuzga I'm using docker to isolate the code (though I know there is still @ the kernel level).

@@michaelbannen5852 what do you mean isolate? It runs in WebContainer that is already isolated.

Hmm. Which provider and model are you using. I for now am using google gemini with it.

Thanks! I kinda start to dislike word Agent :D In reality what it should mean "system that can chase the goal for prolonged time without human intervention and actually get there" Something akin to hype around Devin But I don't think current models can do it, no meter how well you prompt and integrate them. What you said there sounded more like. Make it work better in general :D We want to, lot of work to do there.

@@EduardsRuzga Every tool has its hype or its really usage. AI Agents are a fantastic tool. In the right hand can do very nice things. Its like everything , you know!!! Dont afraid evolution...we should afraid the wrong hands!!!

In local setup or in hosted on website?

@ on te hosted version

Why do you ask? To run local models? I would look at ollama and LM Studio requirements then. If you mean bolt then it even runs on a phone when hosted. I did this video on old decade old Dell laptop with weak AMD GPU that can barely run games and my first recording attempt failed :D

:D This video is not about bolt/oTToDev lol :D

So, for now I gave up a bit. Here is a repo github.com/wonderwhy-er/ClaudeComputerCommander It kinda works and kinda does not, will need to figure out why, asked for help in Antropic github for now Even things that did work in live stream do not work anymore :D While I did have moments when terminal commands worked

Well... I think good example is alphafold. First results from it about things we do not know are what, 1-2 years old. Did world change much? We do not have infrastructure to effect the world with physical discoveries at same speed as we do with digital things. That on itself shows that AI will have friction in adoption with physical things. So, I agree with you, it will do those things. But speed odlf really adopting and applying them in real world will not be as super human