Listening to #ai and #lmm discussion at #ilide I am not convinced. The big elephant in the room: the effects on privacy, copyright, environmental impact and democracy is undersold, the benchmarks, superintelligence and economic impact is oversold. If it is superintelligent, I need to see superintelligent use-cases but I saw only faster horses.
Recent searches
Search options
#LMM
0 posts0 participants0 posts today
A wiecie, że wchodząc na https://duck.ai, możecie pogadać z LLM-ami (w tym z europejskim Mistralem), nie sprzedając diabłu swoich danych, bez logowania i w ogóle?
Już 12 maja 2025 rusza kolejna edycja szkolenia AI Devs, którego autorami są Jakub Mrugalski, Mateusz Chrobok i Adam Gospodarczyk. Jest to edycja "3.0 - RELOADED", która skupia się na tworzeniu agentów AI, które docelowo kursanci integrują z istniejącą infrastrukturą IT.
https://www.aidevs.pl/?ref=z3s
Dzięki szkoleniu dowiesz się jak:
↳ podłączyć LLM-y do własnego źródła wiedzy (np. dokumentacji firmowych usług),
↳ korzystać z baz wektorowych, grafowych i relacyjnych w połączeniu z AI,
↳ optymalizować zapytania pod względem szybkości wykonania, jak i generowanych kosztów,
↳ tworzyć bezpieczne aplikacje, które nie doprowadzą do przypadkowego wycieku wrażliwych danych,
↳ łączyć usługi wielu dostawców LLM i wykorzystywać lokalne modele językowe.
Szkolenie ma formułę kohortową i trwa 5 tygodni. Wszyscy kursanci zaczynają i kończą je w tym samym momencie, co oznacza, że nie będzie to kolejne, odłożone na później szkolenie. Będziesz mieć stały dostęp do trenerów i osób świadczących pomoc dla kursantów, a także - co bardzo ważne - zyskasz dostęp do zamkniętej społeczności pomagających sobie wzajemnie kursantów.
Jeśli obawiasz się, czy aby na pewno podołasz i czy Twoja wiedza jest wystarczająca, aby odnieść korzyści z tego szkolenia, to przetestuj je na własnej skórze. Od daty startu (a nie zakupu) masz 2 tygodnie na rezygnację, bez podania przyczyny i bez ponoszenia jakichkolwiek kosztów. Oznacza to, że możesz za darmo przerobić niemal połowę szkolenia i dopiero wtedy zdecydować, czy zostajesz do końca.
Pełną agendę szkolenia, jak i opinie o nim znajdziesz na poniższej stronie, ale biorąc pod uwagę, ile tysięcy programistów zaufało dotychczas jego organizatorom, z pewnością masz wśród swoich znajomych przynajmniej jednego kursanta AI Devs. Zapytaj go więc o wrażenia :)
https://www.aidevs.pl/?ref=z3s
UWAGA: szkolenie przeznaczone jest dla programistów. Nie jest to nauka pisania sprytnych promptów, czy wykorzystywania gotowych interfejsów webowych. Praca wyłącznie z API - nie musisz być programistycznym ekspertem, ale musisz umieć programować. Rozwiązania no-code nie wystarczą.
P.S. Dopisek RELOADED w nazwie oznacza, że jest to powtórka edycji 3.0 z nową grupą. Jeśli uczestniczyłeś w edycji 3.0, to jest to materiał, który już znasz.
www.aidevs.plAI_devs 3 – Programuj Autonomicznych Agentów i Narzędzia Generative AIW trzeciej edycji AI_devs w pełni wykorzystamy większe możliwości dużych modeli językowych i w połączeniu z programistycznymi umiejętnościami zbudujemy autonomiczne i częściowo autonomiczne systemy, określane mianem "Agentów".
Excellent quote by Dr. Subbarao Kambhampati. Large Mumbling Models.
'I have argued elsewhere that the "reasoning trace" anthropomorphization holds little water and we may have moved from LLMs to LMMs--Large Mumbling Models that have higher accuracy. https://youtube.com/watch?v=CQ5JS3v61Ns&list=PLNONVE5W8PCRbf3WmbcqgXPToJuA2NUfP&t=3786s… )'
youtube.com- YouTubeEnjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.
Microsoft trimming its generative AI sails…
A week ago, analyst TD Cowen revealed that Microsoft had canceled leases "totalling a couple hundred MWs," with "at least two private data center operators across multiple US markets." The report also details how Microsoft "pulled back on converting negotiated and signed Statement[s] of Qualifications (SQQs)," which it added was "the precursor to a data center lease."
Ed Zitron's Where's Your Ed At · Power CutA week ago, analyst TD Cowen revealed that Microsoft had canceled leases "totalling a couple hundred MWs," with "at least two private data center operators across multiple US markets." The report also details how Microsoft "pulled back on converting negotiated and signed Statement[s] of Qualifications (SQQs)," which it added
Replied in thread
First and foremost your health is priority one.
Second you are moving on unknown territory so it is to be expected that certain things make Corners that you do not expect. Third because of the second Point certain steps can take longer, than your ideal plan of execution.
Stay calm, balance your health, then do the rest
The community has put faith in you, I cannot speak for them yet be sure I have faith in you. You shall be able to get a localized version of alt bot running on the server that the community has provided.
Take it Easy relax Balance your health
#LLM and #LMM capability isn't about measuring single models and optimizing those against leaderboards; it's about sustainable progress on the improvement trend. There is a Moore's law analogue in AI capability progress, and every lab needs to keep up with that curve in a sustainable fashion.
Short term hacks are meaningless and are ground to obsolescence in a short amount of time.
If your #AI strategy isn't based on a sustainable progress of capabilities and is only targeting a specific milestone, regardless of whether you meet that milestone or not, you will sink after it.
Getting ready for the upcoming semester's #IoT course here at #ITU -
and doing some quick quiz with #DeepSeek , to see whether the new kid on the thieves' block would pass my exam.
Result: it s about as useless as #ChatGPT. Many factual errors, #hallucinations on key questions of #physics, #sensors, #LoRa, LoRaWAN, general #wireless.
#ThalesUK Demonstrates Versatility of Its #Lightweight Multi-role Missile #LMM Against Simultaneous #Targets
https://armyrecognition.com/news/army-news/2025/thales-uk-demonstrates-versatility-of-its-lightweight-multi-role-missile-lmm-against-simultaneous-targets
@DefenceHQ @BritishArmy
@thalesgroup @ThalesAerospace
#bot #miltodon
Original tweet: https://nitter.nl/ArmyRecognition/status/1879100604242649105
Replied in thread
@EUCommission Better laws before implementing #AI A Law that makes that every use of an AI model must come with transparency, insight in how it works and always with notable an warning that ai is used. And stop #american #lmm #AI #musk #meta #microsoft #copilot #google from intervene in our society. But the most #EU country’s and the eu itself have sold there #IT to #US #BigCo it bring us #fascism before we realise it.
Loke: Digital Road Tax And Driving Licenses Are Here To Stay #automotive #drivinglicense #lmm #roadtax
https://www.lowyat.net/2024/335926/loke-digital-road-tax-and-driving-licenses-are-here-to-stay/
Ukraine Daily Summary - Friday, September 6
Downing missiles while leaving launchers intact is like swatting mosquitoes without closing the window -- As Ukrainian children return to school, Russia launches missiles at educational facilities -- UK to provide Ukraine with $213 million package of air defense missiles -- Russian advance near Pokrovsk stalled in recent days -- Ukrainian forces preempted Russian offensive plans in Kursk -- and more
https://writeworks.uk/~/UkraineDaily/Ukraine%20Daily%20Summary%20-%20Friday,%20September%206/
writeworks.ukUkraine Daily Summary - Friday, September 6Downing missiles while leaving launchers intact is like swatting mosquitoes without closing the window -- As Ukrainian children return to school, Russia launches missiles at educational facilities -- UK to provide Ukraine with $213 million package of air defense missiles -- Russian advance near Pokrovsk stalled in recent days -- Ukrainian forces preempted Russian offensive plans in Kursk -- and more
"AI isn't Intelligent" is the new "Serverless has servers" in that dimwits think it makes them sound smart.
Continued thread
Another great example how #LMM #AI produces extremely confident BS when presented with something (in this case, a rare name) not represented in the training data https://kansasreflector.com/2024/06/22/artificial-intelligence-has-spread-lies-about-my-good-name-and-im-here-to-settle-the-score/
Kansas Reflector · Artificial intelligence has spread lies about my good name, and I'm here to settle the score • Kansas ReflectorArtificial intelligence lies. Everyone knows this by now, of course. Programs such as Chat GPT and Google's "AI overviews" generate nonsense.
#statstab #106 Balancing Type I error and power in linear mixed models
Thoughts: Frequentists will struggle to adopt better modelling practices. Either nominal Type I (random slopes) or realistic power requirements (non-maximal)
https://www.sciencedirect.com/science/article/pii/S0749596X17300013?via%3Dihub
Yesterday I asked both #Google #Gemini and #Microsoft #Copilot to draw me something and I was very specific in my request. It was crystal clear: draw a logo with very specific characteristics.
Neither of them were capable of it. This is what many people are salivating about? Fuck these things! #AI #LMM
Even the #LMM 's are on our side on this one. I had one create a fictional persona of an "average American" and it came up with a cisgendered midwest father of two, mildly conservative, that I could ask questions of. "He" fixes cars.
llama3-70b has internalized some of our more idealistic self-myths, of the US, and I don't think that's a bad thing.
Sobre la mística de las propiedades "emergentes" de la Inteligencia Artificial. Las tendencias místicas del mundo tecnológico no dejan de sorprenderme. #AI #LMM
(10) 
Does generative AI have ‘emergent’ properties?
https://untangled.substack.com/p/does-generative-ai-have-emergent
Untangled with Charley Johnson · 🧠 Does generative AI have ‘emergent’ properties?
#statstab #51 R Functions for Variance Decomposition {varde}
Thoughts: A useful package to get more insight into your mixed effects model.
GitHubGitHub - jmgirard/varde: R Functions for Variance DecompositionR Functions for Variance Decomposition. Contribute to jmgirard/varde development by creating an account on GitHub.