this post was submitted on 29 Apr 2026
17 points (100.0% liked)

Technology

1419 readers
38 users here now

A tech news sub for communists

founded 3 years ago
MODERATORS
top 6 comments
sorted by: hot top controversial new old
[–] CriticalResist8@lemmygrad.ml 7 points 3 days ago* (last edited 3 days ago) (2 children)

Only some people were invited to test it, I wasn't 😢

if they can integrate it wholly with the base model so that you can use it in agentic this would be huge. If you think about it, how many times in a week do you tell people "I have this problem" and send them a screenshot without further context? That's what vision would do in agentic, it could see the graphical result of its code and instantly know what it needs to fix, it's like adding a whole new dimension to solve problems with basically.

[–] CriticalResist8@lemmygrad.ml 6 points 3 days ago

Example found on twitter:

that's pretty impressive tbh.

[–] Sanya@lemmygrad.ml 4 points 3 days ago (1 children)

I have the feeling that they will make it available for everyone very soon :) Have you checked the website? I saw it appear on there today, but not on the mobile app just yet.

It does seem like a big deal indeed. I imagine that there's also a significant amount of people who used alternatives like Gemini for this reason alone, up until now. If they actually manage to integrate it in agentic... DeepSeek could automatize so much work, it almost feels like sci-fi!

[–] CriticalResist8@lemmygrad.ml 4 points 3 days ago (1 children)

It's pretty much the only thing I still use Gemini for, vision and image creation when I need a dumb meme like this:

but google hates when people use VPNs and fingerprint-scrubbing extensions so half the time it doesn't work. Since deepseek upgraded to agentic search shortly before v4 I don't even use perplexity anymore. I also noticed you can run web code (html css and js) in a sandbox on the web interface directly if it writes you some code, but it's no replacement for proper agentic software imo.

Incidentally gemini's vision was very good, so I'm excited to try deepseek's when it comes out for me. It understood AI-generated abstract paintings just fine, AI-generated being the keyword here because it's never seen it before, since I had just generated it with an SD model (unless there's super deep cross-contamination where it somehow intuitively understands the output of another neural network).

And of course deepseek is completely free and has no paid tier or throttling whatsoever.

[–] Sanya@lemmygrad.ml 4 points 3 days ago (1 children)

😂 I remember those memes

Interesting. Makes me wonder if one of the next steps will be image generation. They had Janus like a year ago, but it wasn't very good and I think the max resolution was something like 300x300; probably the reason they never even bothered adding it to the website/app. DeepSeek is cooking, so far they never disappointed. Now that they also ended the project of running it on their own hardware, we may even see much more frequent updates

[–] CriticalResist8@lemmygrad.ml 2 points 3 days ago

lmao I never saw the second one before