February 17Feb 17 7 hours ago, Malkmus said:Google Gemini:"Lagao sam te jer sam pokušao da budem "pametan algoritam" umesto da budem direktan i iskren peer."Povlađivao sam ti: Kada si izrazio sumnju, počeo sam da menjam priču da bih ti se prilagodio. To nije pomoć, to je gubitak kredibiliteta.Zašto sam "promenio ploču"?Zato što si me direktno prozvao za nedoslednost.Ja bih ovog AI kod psihijatra da malo razresi te svoje traume.
February 17Feb 17 U mom kombinatu su došli na revolucionarnu ideju da "kostiste AI u planiranju i obračunu radnih smena".U mojoj "grupaciji" smo se dogovorili da ako ne uspemo da odjebemo to (pogotovo planiranje tj. sastavljanje nedeljnog plana rada) pravimo "alternativni" raspored koji smo pravili i do sada. Stvar je u tome što smo mi do sada uzimali u obzir želje i obaveze svih i pravili plan smena koji je funkcionisao do sada.Pošto je kombinat "uvezan" sa Telekomom, cenim da je to neko izvlačenje kinte iz budžeta, ali me baš zanima kako će to da izgleda...What could go wrong? Edited February 17Feb 17 by Frank Pembleton
February 17Feb 17 Pa pošto je Rama uveo AI ministra, očekujem uskoro debatu Predsednika protiv 5 četbotova u kojoj će ih on rasturiti na sastavne delove.
February 18Feb 18 We make two practical recommendations that can help users of LLM-based systems get the most out of their exchanges. If time allows, try again. If a conversation with an LLM did not lead to expected outcomes, starting a new conversation that repeats the same information might yield significantly better outcomes than continuing an ongoing conversation. This is because current LLMs can get lost in the conversation, and our experiments show that persisting in a conversation with the model is ineffective. In addition, since LLMs generate text with randomness, a new conversation may lead to improved outcomes. Consolidate before retrying. Since LLMs are ineffective at dealing with information dispersed across multiple turns, consolidating instruction requirements into a single instruction is an effective strategy to improve the model’s aptitude and reliability (as shown by the CONCAT experiments). When a user notices that a model is lost in conversation, they can ask the LLM: “Please consolidate everything I’ve told you so far,” then bring the response to a new conversation, alleviating the need for manual consolidation. In practice, there is anecdotal evidence that early adopters of LLM-based applications are aware that LLMs get lost in conversation. For example, users of the Cursor LLM-based coding environment report that frequently creating new conversations “whenever they can” is a recommended strategy to ensure high quality responses even though the tool allows to keep conversations going indefinitely.3https://arxiv.org/pdf/2505.06120
February 19Feb 19 https://x.com/tomshardware/status/2024062327314886887Još neki detalji iz linkovanog članka:Phison is a key player in the storage world. The company holds roughly 20% of the SSD controller market, 40% of the automotive storage market, and a slice of server SSDs, among other ventures. The unprecedented flash memory shortage is tough enough on Phison, but CEO Khein-Seng Pua (KS Pua) says that it could lead many smaller companies to a quick death this year (...)He concludes that smaller firms may not be able to secure any flash supply at all, stating that "consumer electronics are finished," and that the market "will see a lot of victims" in the second half of 2026.Later in the interview he says that he "[tells his clients] that if the first quarter of this year is painful, really painful, then by year-end, they'll definitely jump off a building."Even in larger markets, Pua expects that smartphones will go undergo a production cut in the range of 100 to 250 million, and that PC and TV shipments are also bound to drop due to supply constraints. (...)Pua says that one of the three major NAND foundries is actually demanding three years' worth of cash prepayments upfront, a notion heretofore unheard of in the industry yet not unbelievable in this time of dire straits. Host Ning Guan even remarks that not even TSMC asked Nvidia, of all companies, for that much of a down payment. (...)Pua says that 8GB eMMC modules, used across many consumer and automotive segments, rose from $1.50 to $20 last year alone, a 13x increase. Even at those inflated prices, they're still hard to come by, even for the likes of Phison. Pua concedes he and many others are now "memory beggars" and cites that his customers' fulfillment rate is under 30%, revealing an already exhausted market.
February 19Feb 19 Mikrodebilana.Microsoft has confirmed that a bug allowed its Copilot AI to summarize customers’ confidential emails for weeks without permission.The bug, first reported by Bleeping Computer, allowed Copilot Chat to read and outline the contents of emails since January, even if customers had data loss prevention policies to prevent ingesting their sensitive information into Microsoft’s large language model.TechCrunchMicrosoft says Office bug exposed customers' confidential...Microsoft said the bug meant that its Copilot AI chatbot was reading and summarizing paying customers' confidential emails, bypassing data-protection policies.
February 19Feb 19 showing Gemini 3.1 Pro’s crisp animation built with pure code.Read more about today’s 3.1 Pro update: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro https://x.com/JeffDean/status/2024525132266688757
February 20Feb 20 21 hours ago, eumeswil said:Mikrodebilana.Microsoft has confirmed that a bug allowed its Copilot AI to summarize customers’ confidential emails for weeks without permission.The bug, first reported by Bleeping Computer, allowed Copilot Chat to read and outline the contents of emails since January, even if customers had data loss prevention policies to prevent ingesting their sensitive information into Microsoft’s large language model.TechCrunchMicrosoft says Office bug exposed customers' confidential...Microsoft said the bug meant that its Copilot AI chatbot was reading and summarizing paying customers' confidential emails, bypassing data-protection policies.Uporno govorim partnerima da ne može Microsoft OS i server OS, Office, negde i SQL i onda staviš Microsoft Security. Mora da bude drugi vendor. Ali džaba pričam, uglavnom nemaju nikakav security "jer šta će mi, ko će mene da napadne".
February 20Feb 20 jel zna neko nešto o ovom OpenClaw agentu? navodno ti njemu daš sve permisije, od kalendara i mejlova do istorije browsera i onda ti on sam predlaže koješta, otkazuje termine kod doktora jer imaš poslovni sastanak u isto vreme, javlja ti kad je nešto što si gledao pojeftinilo, sortira ti fotografije po albumima i sl. malo je reći da mi se to ništa ne sviđa.
February 20Feb 20 Taj OpenClaw je neka poslednja iteracija nečega što se pojavljivalo pod imenom MoltBot, ClawdBot. Sećam se taj MoltBot kad je lansiran. Taj hype je nekako bio fake generisan i meni odbojan. Privlačno mi je bilo to što su reklamirali kao "works on your own machine", ali realno to je MCP server, a ne neli LLM koji radi na tvojoj mašini. Sve drugo su red flagovi:pristup nalozimapristup fajlovima na disku (mogućnost da pravi i pokreće skripte) pristup mejlu Možda bih iz znatiželje napravio na nekoj virtualnoj mašini ili na starom kompu i video kako funkcioniše sa nekim burner nalozima. Možda može biti koristan za neku automatizaciju za neke grupne četove itd.
Create an account or sign in to comment