technology

<p>LLMs totally choke on long context because of that O(n2) scaling nightmare. It’s the core scaling problem for almost all modern LLMs because of their self-attention mechanism.</p><p>In simple terms, for every single token in the input, the attention mechanism has to look at and calculate a score against every other single token in that same input.</p><p>So, if you have a sequence with n tokens, the first token compares itself to all n tokens. The second token also compares itself to all n tokens… and so on. This means you end up doing n*n, or n^2, calculations.</p><p>This is a nightmare because the cost doesn’t grow nicely. If you double your context length, you’re not doing 2x the work; you’re doing 2^2=4x the work. If you 10x the context, you’re doing 10^2=100x the work. This explodes the amount of computation and, more importantly, the GPU memory needed to store all those scores. This is the fundamental bottleneck that stops you from just feeding a whole book into a model.</p><p>Well, DeepSeek came up with a novel solution to just stop feeding the model text tokens. Instead, you render the text as an image and feed the model the picture. It sounds wild, but the whole point is that a huge wall of text can be “optically compressed” into way, way fewer vision tokens.</p><p>To do this, they built a new thing called DeepEncoder. It’s a clever stack that uses a SAM-base for local perception, then a 16x convolutional compressor to just crush the token count, and then a CLIP model to get the global meaning. This whole pipeline means it can handle high-res images without the GPU just melting from memory activation.</p><p>And the results are pretty insane. At a 10x compression ratio, the model can look at the image and “decompress” the original text with about 97% precision. It still gets 60% accuracy even at a crazy 20x compression. As a bonus, this thing is now a SOTA OCR model. It beats other models like MinerU2.0 while using fewer than 800 tokens when the other guy needs almost 7,000. It can also parse charts into HTML, read chemical formulas, and understands like 100 languages.</p><p>The real kicker is what this means for the future. The authors are basically proposing this as an LLM forgetting mechanism. You could have a super long chat where the recent messages are crystal clear, but older messages get rendered into blurrier, lower-token images. It’s a path to unlimited context by letting the model’s memory fade, just like a human’s.</p> <small class="notice" x-post-type-data="None"> Takahe has limited support for this type: <a href="https://lemmy.ml/post/37805263">See Original Page</a> </small>

<p>If anybody has a better non-Forbes link, thank you. I hadn’t heard about this anywhere else apart from a podcast. Aren’t there any petitions or pressure campaigns against google’s overreach?</p> <small class="notice" x-post-type-data="None"> Takahe has limited support for this type: <a href="https://mander.xyz/post/45949623">See Original Page</a> </small>

<p>In my view, this is the exact right approach. LLMs aren’t going anywhere, these tools are here to stay. The only question is how they will be developed going forward, and who controls them. Boycotting AI is a really naive idea that’s just a way for people to signal group membership.</p><p>Saying I hate AI and I’m not going to use it is really trending and makes people feel like they’re doing something meaningful, but it’s just another version of trying to vote the problem away. It doesn’t work. The real solution is to roll up the sleeves and built an a version of this technology that’s open, transparent, and community driven.</p> <small class="notice" x-post-type-data="None"> Takahe has limited support for this type: <a href="https://lemmy.ml/post/41648644">See Original Page</a> </small>

<p>Lots of venture capital money behind it. I wonder how quickly the enshitification will begin.</p> <small class="notice" x-post-type-data="None"> Takahe has limited support for this type: <a href="https://lemmy.world/post/41626455">See Original Page</a> </small>

<p>I don’t understand subscribing to music. Maybe it’s just my age, but this isn’t the &#x27;90s where you hear a track you like and that one song is going to run you $20 at Tower Records. I like a song, I pay $1.29 and then it’s stored locally. Also cuts way down on data usage while driving. I struggle to get anywhere close to my 5GB data allowance.</p><p><p>After a dozen years of keeping subscription prices stable, Spotify has issued three price hikes in 2.5 years.</p><p>Spotify informed subscribers via email today that Premium monthly subscriptions would go from $12 to $13 per month as of users’ February billing date. Spotify is already advertising the higher prices to new subscribers.</p><p>Although not explicitly mentioned in Spotify’s correspondence, other plans are getting more expensive, too. Student monthly subscriptions are going from $6 to $7. Duo monthly plans, for two accounts in the same household, are going from $17 to $19, and Family plans, for up to six users, are moving from $20 to $22.</p><p>Spotify’s Basic plan, which is only available as a downgrade for some Premium subscribers and is $11/month, is unaffected.</p><p>For years, Spotify subscribers enjoyed stable prices, but today’s announcement marks Spotify’s third price hike since July 2023. Spotify last raised prices in July 2024. Premium individual subscriptions went from $11 to $12, Duo subscriptions went from $15 to $17, and Family subscriptions increased from $17 to $20.</p></p> <small class="notice" x-post-type-data="None"> Takahe has limited support for this type: <a href="https://beehaw.org/post/24258631">See Original Page</a> </small>

<p>&quot;On Monday, OpenAI released data estimating that 0.15 percent of ChatGPT’s active users in a given week have conversations that include explicit indicators of potential suicidal planning or intent. [...] OpenAI also estimates that a similar percentage of users show heightened levels of emotional attachment to ChatGPT [...]&quot;</p><p><a href="https://arstechnica.com/ai/2025/10/openai-data-suggests-1-million-users-discuss-suicide-with-chatgpt-weekly" rel="nofollow" class="ellipsis" title="arstechnica.com/ai/2025/10/openai-data-suggests-1-million-users-discuss-suicide-with-chatgpt-weekly"><span class="invisible">https://</span><span class="ellipsis">arstechnica.com/ai/2025/10/ope</span><span class="invisible">nai-data-suggests-1-million-users-discuss-suicide-with-chatgpt-weekly</span></a></p><p><a href="/tags/news/" rel="tag">#news</a> <a href="/tags/technews/" rel="tag">#TechNews</a> <a href="/tags/technology/" rel="tag">#technology</a> <a href="/tags/ai/" rel="tag">#AI</a> <a href="/tags/mentalhealth/" rel="tag">#MentalHealth</a></p>