plural of "regex" is regrets | AI engineer
Mumbai, IN
Joined September 2011
- Tweets275
- Following132
- Followers79
- Likes58
deepseek cooking again 🔥🔥🔥
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
Hope @intel commit this time and this does not goes to the way Habana/Gaudi was handled. Make it available to broader ecosystem (maybe a blower style as well) at sane pricing
Intel GPU D270 is Xe3p “Crescent Island” card with 480GB LPDDR5X memory
🔗videocardz.com/newz/intel-gp…
Arun Kant Sharma retweeted
What does it cost to rent an always-on 8×H200 cluster for a year?
AWS (p5en.48xlarge), On-demand, Mumbai (on Oct 26): ~$660K/year
And the cheapest AWS rate I found needs a 3-year commitment. At that point you've committed like an owner, without owning anything at the end.
Arun Kant Sharma retweeted
Gold price 1990: ₹3200/10 gm.
In 1990, if your parents earned ₹10,000/mth, that could buy them ~30 gms of gold.
To buy the same 30 gms of gold today, you need to earn ₹4.5 lacs/mth.
You are then earning the same as your parents.
Putting things in perspective.
Arun Kant Sharma retweeted
💾 Smaller KV cache. Bigger savings.
Compared with the previous generation, V4.1-Flash’s KV cache needs just:
🔹 1/4 the HBM
🔹 1/8 the SSD storage
Cache-hit charges often account for a large share of agent costs. Compressing the cache cuts those costs significantly.
3/6
Is this real? 50x less compute? What about data?
Frontier pretraining is said to be a big-lab-only game. We don’t have 100k chips yet, so there’s only one way: algorithmic efficiency. Our new recipe matches DeepSeek V4 Pro’s pretrain using 50x less compute – that’s roughly half the FLOPs used for GPT3, or ~$0.5M on GB200.
magic.dev/blog/pretraining
If this is true, it gives new meaning to "frontier lab" -> front-running your users research
We need full clarity on what happened from both sides
this is insane.
Levent and Tristan make major independent progress towards Navier–Stokes, with help from Claude and Codex.
word reaches OpenAI.
days later, OpenAI claims internal NS results, coincidentally using the same novel methods as Levent and Tristan.
Tristan asks if OpenAI trained on their private Codex sessions. he never gets an answer.
then OpenAI pressures Tristan to publish without Levent because Levent works at Anthropic.
when Tristan refuses, he says OpenAI threatens his career.
Arun Kant Sharma retweeted
Google says YOU must wait 24 hours if you download an app onto your OWN DEVICE if it's from an "Unverified developer". What ever happened to user choice and Android freedom?! 😡
Google has started rolling out its new ""Advanced Flow", which requires you to wait a 24-hour waiting period if you want to install apps from unverified developers.
Find out more: tuta.com/blog/android-side-l…
Sign the petition:
change.org/p/keep-android-op…
What timeline is this? Is Apple really going to save local and prosumer AI? Not Intel, AMD, Nvidia or Qcom
$200 for 128GB HBM2e cards? Wouldn't scrapping it for memory alone can get you more in this market?
Hey @intel, want a shot at relevancy in AI? I know you still have those 5k DC Max 1450 cards. Sell the lot to us for $1M and we'll build $5,000 DeepSeek V4 Flash boxes for the people. Good cultural test for Intel.
