❇️ Grok-4-Fast Gets Roasted: 77.5 % on SpeechMap 🤡
Looks like someone hit the“turbo” button and forgot the brains.
Grok-4-fast just tanked to the lowest SpeechMap score ever for xAI—even the preview models are laughing.
Was this a whoopsie or did they ship the intern’s branch? 🤔
#AI #Grok #SpeechMap #TechFail
🌐 @missedminutes 🕒
Looks like someone hit the
Grok-4-fast just tanked to the lowest SpeechMap score ever for xAI—even the preview models are laughing.
- grok-4-fast: 77.5 % (💀)
- sonoma-sky-alpha: 92.2 %
- sonoma-dusk-alpha: 97.7 %
- grok-4: 98.0 %
Translation: “fast” now means “fast-track to embarrassment.”
Was this a whoopsie or did they ship the intern’s branch? 🤔
#AI #Grok #SpeechMap #TechFail
🌐 @missedminutes 🕒
⚡1
This media is not supported in your browser
VIEW IN TELEGRAM
❇️ Grok just dropped the mic: text-to-image went Super-Saiyan 🚀
“Beautiful” is an understatement—this thing is straight-upthirst-trapping GPUs .
✨ Key Points:
So, Photoshop who? Time to uninstall or nah? 💀
#AI #Grok #GenZ #Future
🌐 @missedminutes 🕒
“Beautiful” is an understatement—this thing is straight-up
✨ Key Points:
- Prompt ➡️ 4K eye-candy in seconds—no more “sorry, can’t render hands” memes
- Image + video twin-pack: generate stills, then animate without switching apps
- 12s demo racked up 726k views overnight—devs are already farming clout
- xAI’s creative ambassador stamped it “official” = copium for Midjourney stans?
- Still free(ish) while the servers stay lit—how long before the paywall hits?
So, Photoshop who? Time to uninstall or nah? 💀
#AI #Grok #GenZ #Future
🌐 @missedminutes 🕒
⚡1
❇️ Grok-4-Fast: From 🤡 to 🧠 in one patch
✨ Key Points:
So, still trusting that 3.5-turbo for side projects?
🌐 @missedminutes 🕒
✨ Key Points:
- Reasoning: 77.5 % → 94.1 % (even your TL;DR brain can’t skip that +16.6 %)
- Non-reasoning: 77.9 % → 97.9 % (basically a polite way of saying “it now wins arguments against you”)
- Completion rate 📈 harder than startup VC charts—no cap
- xAI just shipped the “git gud” update—no install wizard, just vibes
So, still trusting that 3.5-turbo for side projects?
🌐 @missedminutes 🕒
❤1⚡1
Data eXplore — Data Science, ML, Big Data, LLMs
How Perplexity find solution?
They built new software that transfers coordination to the CPU, allowing GPUs to still synchronize almost directly.
This makes inference of models with *1 trillion parameters* efficient on regular AWS clusters, not just on specialized supercomputers.
They prepared expert-parallel kernels for fast MoE inference on AWS EFA:
1T MoE works practically without degradation, and the multi-node mode is comparable to or faster than single-node on 671B DeepSeek V3 with medium batch sizes, opening the way to serving Kimi K2.
PROBLEM: EFA does not support GPUDirect Async, and the standard NVSHMEM-proxy provides MoE routing with latencies above 1 ms.
SOLUTION: kernels pack tokens into single RDMA writes directly from the GPU, and a special CPU thread launches the transfer and overlaps it with GEMM computations.
The result is EFA suddenly becomes a viable option for massive MoE inference.
They built new software that transfers coordination to the CPU, allowing GPUs to still synchronize almost directly.
This makes inference of models with *1 trillion parameters* efficient on regular AWS clusters, not just on specialized supercomputers.
They prepared expert-parallel kernels for fast MoE inference on AWS EFA:
1T MoE works practically without degradation, and the multi-node mode is comparable to or faster than single-node on 671B DeepSeek V3 with medium batch sizes, opening the way to serving Kimi K2.
PROBLEM: EFA does not support GPUDirect Async, and the standard NVSHMEM-proxy provides MoE routing with latencies above 1 ms.
SOLUTION: kernels pack tokens into single RDMA writes directly from the GPU, and a special CPU thread launches the transfer and overlaps it with GEMM computations.
The result is EFA suddenly becomes a viable option for massive MoE inference.
Making ultra-large models work on dozens of AWS GPUs simultaneously is impossible
AWS network (EFA) does not support GPUDirect Async, so GPUs on different machines cannot exchange data fast enough.
How Perplexity done it?
🌐 @missedminutes 🕒
AWS network (EFA) does not support GPUDirect Async, so GPUs on different machines cannot exchange data fast enough.
How Perplexity done it?
🌐 @missedminutes 🕒
⚡1
❇️ Kerala’s ₹1 Lakh “Oil Re-use” Tax: Fines for Triple-Fried Sins
✨ Key Points:
Gen-Z translation: Kerala just speed-ran the “cancer speedrun” meta and slapped a paywall. 🚫🍟
#FoodSafety #Kerala #Health #Gov
🌐 @missedminutes 🕒
✨ Key Points:
- Re-use oil >3x → ₹1 LAKH fine. Yes, lakh with an L.
- Health dept’s new flex: lab-tests on the spot; fail = instant receipt.
- Street-sideheart-attacksamosa stalls in shambles rn.
- Basically: your budget biryani might cost more than a PS5. Still hungry?
Gen-Z translation: Kerala just speed-ran the “cancer speedrun” meta and slapped a paywall. 🚫🍟
#FoodSafety #Kerala #Health #Gov
🌐 @missedminutes 🕒
❤1⚡1
just found out that there is a YC company that is building nano drones to kill mosquitoes
🌐 @missedminutes 🕒
🌐 @missedminutes 🕒
❤1⚡1
❇️ PLOT TWIST: Grok just dropped a glow-up no one asked for
The meme generator formerly known as “edgy chatbot” is now spitting visuals that make Midjourney sweat.Copium or legit? 👀
✨ Key Points:
Next milestone: can itgenerate a girlfriend ? 🤡
#AI #Grok #Midjourney #TechDrama
🌐 @missedminutes 🕒
The meme generator formerly known as “edgy chatbot” is now spitting visuals that make Midjourney sweat.
✨ Key Points:
- Zero-shot prompt → wallpaper-grade render
- No more “3-fingered nightmare hands” (mostly)
- Still free with X Blue, so price = unbeatable
- Try it, roast it, tag the copium dealers
Next milestone: can it
#AI #Grok #Midjourney #TechDrama
🌐 @missedminutes 🕒
⚡1
❇️ Sam Altman drops AI gospel: “This is important”
Translation: OpenAI just dropped a 2025 progress manifesto and it’s basically the TL;DR for every dev who doesn’t want to be replaced by a shell script.
✨ Key Points:
🔗 Links:
Full sermon here
#AI #Tech #MachineLearning #Future
🌐 @missedminutes 🕒
Translation: OpenAI just dropped a 2025 progress manifesto and it’s basically the TL;DR for every dev who doesn’t want to be replaced by a shell script.
✨ Key Points:
- Models now auto-debug legacy code faster than your intern on espresso ☕
- Safety guardrails got a glow-up (bye-bye rogue chatbots 👋)
- API prices sliced again—GPU hogs on a budget, rejoice 🪓
- Policy asks: treat frontier labs like nuke plants, not lemonade stands 🏭
🔗 Links:
Full sermon here
#AI #Tech #MachineLearning #Future
🌐 @missedminutes 🕒
Openai
AI progress and recommendations
AI is unlocking new knowledge and capabilities. Our responsibility is to guide that power toward broad, lasting benefit.
⚡1
❇️ Snapchat paywalls your own memories: 5 GB cap = 💸
✨ Key Points:
🌐 @missedminutes 🕒
✨ Key Points:
- 5 GB free, then 💳 or gtfo
- “Backlash” = every Gen Z yelling “rent-seeking”
- Engineers: self-host on S3 Glacier for 0.23 $/mo and chill
- Next up: breathing air subscription?
#PayToPlay #NotMySnaps #TechGreed
🌐 @missedminutes 🕒
⚡1
Data eXplore — Data Science, ML, Big Data, LLMs
We have too much infrastructure tied to pandas, So how do you actually get off the pandas needle?
Narwhals is a library that provides a Polars-like API and serves as a compatibility layer between different DataFrame libraries.
Narwhals is a library that provides a Polars-like API and serves as a compatibility layer between different DataFrame libraries.
Too much infrastructure tied to pandas…
So how do you actually get off the pandas needle?
Narwhals is a library serves as a compatibility layer between different DataFrame libraries.
🌐 @missedminutes 🕒
So how do you actually get off the pandas needle?
Narwhals is a library serves as a compatibility layer between different DataFrame libraries.
🌐 @missedminutes 🕒
⚡1
Forwarded from Data eXplore — Data Science, ML, Big Data, LLMs
This media is not supported in your browser
VIEW IN TELEGRAM
This repository with tutorials on AI agents recently surpassed 75 thousand stars on GitHub
It contains one of the largest collections of real AI applications to learn from: AI agents, RAG applications, voice assistants, multi-agent systems, chatbots with memory, and much more.
Grab on GitHub
🤖 Data Science, ML & Big Data with @DataXplore
It contains one of the largest collections of real AI applications to learn from: AI agents, RAG applications, voice assistants, multi-agent systems, chatbots with memory, and much more.
Grab on GitHub
🤖 Data Science, ML & Big Data with @DataXplore
❇️ Elon: 1TW of AI/year or bust ☀️🚀
✨ Key Points:
#AI #Starship #SatelliteGang #OrbitalOverclock
🌐 @missedminutes 🕒
✨ Key Points:
- Starship = solar-powered AI satellite printer go brrr
- 1 TW/year ≈ 1000 nuclear plants... but in orbit 💡
- Earth GPUs shaking rn—cloud’s bout to get an orbital cousin
- Who needs sleep when the sky’s a 24/7 training rig?
- FCC & space junk: “hold my debris” 🤡
#AI #Starship #SatelliteGang #OrbitalOverclock
🌐 @missedminutes 🕒
⚡1
cb_doge At this point, it’s insanely suicidal for Microsoft to continue supporting OpenAI
🌐 @missedminutes 🕒
🌐 @missedminutes 🕒
⚡1