... could have. And I still don't have a clean answer on the verification-method question, because the honest version is that I don't yet know what I'd do differently with a government-ID-checked 16 versus a self-declared one. I'd rather say that than write a paragraph pretending I've resolved it. The thing I am confident about is narrower. The signal is sparse

... exists in order not to do. One more, and it is a correction. In update #2 I wrote that values gives you "exactly the schema you asked for and nothing else". That was not true when the schema was sparse. I declared a single field on a delivery note and got four more back that I never asked for, with a declared count of 44. The thinner the schema, the more the model added whatever else it could ...

... paper argues that they still lack the abductive jump needed to originate new foundational hypotheses. Given Einstein's equivalence principle, a model might derive much of the later mathematics. The harder step is inventing that principle when observations are sparse. Why this matters This is not an argument that AI cannot contribute to research. It is a warning that automated proof, experimentation, and optimization are not the same as genuine scientific invention. The proposed path forward is not only larger ...

... Parameter count alone has become a bad shortcut for explaining conditional models. If a system uses experts or sparse memory, I want four numbers: total parameters, parameters active per token, routing or retrieval overhead, and measured throughput on named hardware. Without those, a large number can describe storage capacity while readers assume every parameter contributes to every token. That muddles both cost and capability. Which number do you wish model announcements reported by default ...

... goal: if you want raw eyeballs go Tue-Thu, but for a badge/ranking with a modest following a weekend can punch above its weight. Comment from Sabber Ahamed(@sabber_ahamed): @busmark_w_nika   @abdullah_javaid3  the invitation vs favor framing is the most useful line in this whole thread. i caught myself drafting DMs that were basically upvote asks in a nicer outfit. what actually worked was showing up in threads like this one for weeks with nothing ... ... most important thing is to launch as soon as you can and keep iterating to improve, and keep launching Especially with how fast everything is moving with AI, waiting for a perfect time to launch will just let competitors

... Hugging Face. That freedom lets us explore efficient architectures larger labs may consider too experimental. With Metis-1.6, we re also preparing a paper on our new MoRE architecture, Mixture of Recursive Experts. Without getting too technical, MoRE combines sparse expert selection with recursive computation. The goal is to give the model substantially more reasoning capability than its raw parameter count would suggest while keeping inference fast and inexpensive because only a fraction of the model is active at once ...

... three degrees out. Comment from Rahil Pirani(@rahilpirani): @djbutterrock_goulet   That's what the ranking is for. Direct matches always score above connected memories, so the right answer comes first. The graph adds context when direct evidence is sparse - it doesn't pile on top of a clear answer. Comment from Rahil Pirani(@rahilpirani): That's the direction. The hops parameter is already configurable per MCP call. Next step is surfacing it as a user-facing setting rather than ...

... assistant set and update a chart s description when editing it. Added hover tooltips, a crosshair, and clearer legend and segment interactions to the AI usage charts in billing settings. Fixed vertical bars overlapping on time-series charts with sparse, clustered dates. Fixed setup for Fivetran file connectors like Google Drive so Magic Folder connectors create correctly. Made chat handle files that can no longer be retrieved more gracefully instead of surfacing a generic error. Comment from Mustafa Arian(@thenameisarian ...

... achieves  71.55% zero-shot top-1 accuracy at 84.0% token sparsity  on SigLIP, retaining 92.4% of dense accuracy without modifying any weights. Architecture B (Extreme Wall-Clock Speedup):  Physically eliminates stationary tokens before the encoder. With sparse manifold distillation, it reduces 1792p SigLIP 2 inference from 678 ms to 11.9 ms a  55.80 wall-clock speedup  at 97.37% embedding fidelity. LLM Ablation:  Characterises the architectural boundaries of applying similarity-gated bypass to autoregressive language ...

... intel with time-based context - the kind of specificity I desperately wanted when planning trips but couldn't find anywhere. Known issues: UI is rough, mobile experience needs work, data coverage is limited to two cities, some neighborhoods have sparse data. This is very much beta. What I'm looking for: Feedback on usefulness vs

... role or company on LinkedIn. The platform gives you a lot of algorithmic juice around that moment, so if you re strategic you can get a lot of visibility from it. Early on, this stuff matters, those sparse likes and comments give you some social signal. 3. LinkedIn We re a meeting software tool, so we ve pushed hard on LinkedIn. It makes sense for our category. I ve found personal channels almost always outperform the brand account ... ... Definitely got a bunch of low quality users, there was one AI listing site we used free credits to test their PPC, was just bot traffic Comment from Johanna John(@johanna_john1): Loved the honestly about newsletter traffic vs

... Solo builder here, won't say much about the product yet but happy to talk shop on the plumbing. Quick version of what I'd tell my past self: Auth - Supabase, Clerk, Auth0, pick one. An afternoon of setup vs. days of building it badly. I tried the days version first, do not recommend. Permissions - this is the sneaky one. Most people check "can this user see this?" in their app code. Way better to push the rule into ... ... tattoos or mockups) Tech-wise I m using: Flutter for the mobile app Firebase (Firestore + functions) OpenAI + Replicate for image generation (testing multiple models right now like DreamShaper, Flux, Ideogram) Biggest challenge so far has been balancing quality vs

... search against user queries, pass top matches to an LLM for relevance filtering. It worked good, but not great. Here's why: Problem 1: Facts don't embed well "User streams on Tuesdays" or "Audience is mostly Brazilian" are sparse semantic content. Embeddings need richer context to match meaningfully. Problem 2: Every query required an LLM pass Just to filter results. Expensive and slow. Problem 3: Reducing everything to facts loses information Tweets, conversations, uploaded docs all flattened into ...

... handle hybrid recall or scale gracefully. Enterprise teams need auditability + efficiency, not black-box memory. What GraphBit s Data Layer does differently Unified Retrieval One interface across pgvector, Qdrant, Weaviate, Elastic. No lock-in. Hybrid Search Dense + sparse + graph edges for accuracy & fewer hallucinations. Versioned Memory Short-term, long-term & time-scoped recall with audit logs. Targeted Caching Hot responses + tool I/O caching to cut infra spend. Real-time Ingest CDC streams keep data fresh within seconds. Privacy ... ... doesn t just orchestrate- it scales, predictably. Would love your feedback: What s the biggest data pain you ve faced in scaling AI agents? Try it here: @GraphBit Musa Comment from Musa Molla(@musa_molla): @imrul_kayes10  Freshness vs

... affiliate content to boost revenue, right? Good point. Although we see monetization as a key hook for early adopters struggling to generate content they can monetize. I could own a Shopify Store and generate a comparative research report on competitor products and use that to drive sales for my own store... It's a way to turn content into commerce, or even just to do product research for yourself. RE: Your video speaks to "how hard ... ... please, please do me! https://futurezapp.com Comment from Alok Jain(@alokjain): @steveb   The target for this product are qualitative researchers. With the use of AI in this space, data privacy is top of the mind for people. The competitors

... really impressed me. Simple to use, the delightful format of briefing people, only a positive experience. The only downside is that I cannot see the whole call transcript, which is crucial for me. Are there any free/cheap AI alternatives that makes summaries of answers and saves transcripts at the same time? Comment from steve beyatte(@steveb): Yapz being free is certainly a plus but I'm with you, it's weird that they don't show you the transcript ... ... also feels like it's way too early to send to real customers (product feels quite sparse