Isn’t DeepSeek OCR basically just expanding the context window? Does it have anything to do with cross session memory?
2025-10-25 21:28:54
1
RoboLawyer-tm :
place to keep dbs is in the browser or in the filesystem - every time an llm looks at text it chunks and embeds it so save it!
2025-10-25 15:21:59
2
Jon Capriola :
Why are all LLMs run unauthenticated by default
2025-10-26 16:42:08
0
Michael McBrayer :
Spaces in perplexity has been life changing for this—having a static thread I can save a link and collaborate and do account or topic specific research I can return to a d evolve over time. The space has a prompt and the ability to add docs and links specific to the project or business outcome —that in tandem with the comet browser was a perfect bridge between real time browsing and getting web based data back into a running thread. Of course this was amazing until we were banned from using it last week and I lost access to everything I’ve been working on since March in my work machine 😂
2025-12-20 01:00:11
1
Sean :
I'm convinced the image encoding thing is going to be an enormous red herring. it's going to be latent space representations all the way
2025-10-27 16:35:15
1
DogsAreTheBest :
when she's super smart and super beautiful 👍 Have a great life, queen.
2025-10-25 21:22:24
0
sailor from the future :
I think this approach leads nowhere because the more the conversation grows, the more it loses sight of what is really important.
2025-11-05 18:13:11
1
BamJamwoo :
as chatGPT should. every little thing means something
2025-10-28 19:36:25
0
kiwi_feathers :
i am doing chunking too. i did think about what deepseek is doing but that meant another prompt and another run on AI. so left it at chunk but know the other way is better
2025-10-26 07:03:39
1
Frankly Q :
Or ask them ? “I know its iver priced but so are alot of things. What get you pumped up about an overpriced bed.” Oh you havent been sleeping, oh the baby, oh the new manager gives you extra work cause you finsh before others” “”— sleepless overworked underpaid — is the point of his grief 👊✊
2025-10-27 07:03:35
1
Ladderly.io -Land a Coding Job :
Very nice
2025-10-26 02:50:22
1
biryanidreams1 :
I haven’t dived deep into the details of memory but the deepseek approach sounds so reasonable I wonder why didn’t open ai take that approach to start?
2025-10-26 14:53:29
0
DubSí :
Compression baby
2025-10-27 08:43:24
1
joshuadarlingtonm :
Nice
2025-10-26 21:50:38
1
Against The Norm :
Thank you for the explanation.
2025-10-26 03:12:44
1
The Lonely Entrepreneur :
The issue is a lot of the conversation doesn’t lead to solutions, and are often hallucinated or wrong. Try to remember and reason from these don’t help, it’s much easier to just use a MD file to define what good look like and keep using that.
2025-11-30 10:12:54
0
Valentin :
not 100x but more 10x (which is awesome since it’s theorically impossible according to Shanon information theory)
2025-10-25 23:26:59
2
Super Intelligent AI :
Excellent coverage
2025-10-25 22:40:08
1
AI For Humans :
Deepseek decided to dial into totally new ways of approaching AI and I’m here for it.
2025-10-26 23:47:58
10
NoomNoo :
I use the same conversion in the same kind of topic with DeepSeek or Qwen. The summarization is fast without the problem of memory
2025-11-05 19:18:44
0
E:P:i RE🔻OLUTION \-_-/ :
🥰
2025-10-27 04:04:39
0
• EVOCATIV3 AI • :
… 🧠 …
2025-10-25 23:20:12
0
To see more videos from user @aistartupfren, please go to the Tikwm
homepage.