Thomas Michael

4.4K posts

Thomas Michael

Thomas Michael

@reThomasMichael

Mathematician, Data Scientist, AI consultant

Berlin, Deutschland Katılım Kasım 2021
353 Takip Edilen172 Takipçiler
Pedro Domingos
Pedro Domingos@pmddomingos·
@AndreasSteno Dangerous ignorance. Without continuing American help Ukraine would quickly lose.
English
106
0
153
19K
Pedro Domingos
Pedro Domingos@pmddomingos·
Dear Europe: if Iran is not your problem, Ukraine is not America’s.
English
3.4K
810
7.4K
598.2K
John Rain
John Rain@johnthenoticer·
Germany has 39,000 Japanese citizens; only two of them were suspected of violent crimes in 2023. By contrast, out of 25,000 Algerians, 1,729 were suspected of committing a violent crime in 2023.
John Rain tweet media
English
1.2K
7.2K
43.1K
41.2M
Thomas Michael
Thomas Michael@reThomasMichael·
@HeyCharlieFirpo Diese Aversion gegen Elektroautos wirkt mehr und mehr dümmlich je besser die Elektroautos werden. In ein paar Jahren werden Verbrennerfans so wirken wie die, die heute den VW-Käfer zurück wollen.
Deutsch
0
0
0
5
Charlie Firpo
Charlie Firpo@HeyCharlieFirpo·
BILD und andere Medien titeln heute groß: "Große Auswertung: Immer mehr Vielfahrer steigen aufs E-Auto um" Die "große Auswertung" der HUK gibt nur leider etwas ganz anderes her. 😂 Ein Mini-Thread: 1/6
Charlie Firpo tweet media
Deutsch
387
780
4.8K
532.7K
Thomas Michael retweetledi
Rohan Paul
Rohan Paul@rohanpaul_ai·
Unitree just introduced the A2 Stellar Hunter. How do you even fight an army of these? Like what will be the success rate of a bunch of regular soldiers (or for that matter even special ops) vs these? Their lightest and fastest industrial robot dog yet. - ~37kg - ~20km range - Built for real-world industrial use - can balance on one or two limbs. - It can climb 1m obstacles and run at speeds up to 5m/s and overcome steep terrains. - can handle an adult man jumping on its back while moving. - can perform all kinds of flips. It uses an ultra-wide LiDAR 3D perception system to adapt to its surroundings. Future wars are going to be insane.
Unitree@UnitreeRobotics

Unitree Introducing | Unitree A2 Stellar Hunter 🤩 Total weight: ~37kg | Unloaded range: ~20km Lighter, Stronger and Faster. Engineered for Industrial Applications.

English
2
7
19
5.7K
Thomas Michael
Thomas Michael@reThomasMichael·
Oh my god, running the new open source model by openai locally on my computer is really strangely cool!
English
0
0
1
24
Thomas Michael retweetledi
Science girl
Science girl@sciencegirl·
According to a World Shipping Council report, up to 1,382 shipping containers are lost at sea each year
English
107
92
465
77.8K
Thomas Michael retweetledi
Science girl
Science girl@sciencegirl·
a noise in the roof in Australia 📹 ti. inthewild
English
802
1.4K
9.2K
871.3K
Thomas Michael retweetledi
Igor Sushko
Igor Sushko@igorsushko·
Elite Russian paratrooper celebrates Airborne Forces Day.
English
251
272
1.9K
193.2K
Thomas Michael retweetledi
Aviation
Aviation@xAviation·
Lightning seen from an airplane window!
Aviation tweet media
English
9
131
1.4K
65.7K
Thomas Michael retweetledi
Rohan Paul
Rohan Paul@rohanpaul_ai·
This paper will change how we think about LLM inferencing. 🔥 An alternative to Chain-Of-Thought, instead nspired by how the human brain utilizes distinct systems for slow, deliberate planning and fast, intuitive computation. Gives us super fast reasoning vs SOTA LLMs with just 1,000 training examples and a 27mn param model. Unbelievable how a tiny model from a tiny lab of Tsinghua + deep thinking, gets 40% on ARC-AGI and tear into complex sudoku and maze puzzles. 🤯 It removes token-by-token chain-of-thought generation. Instead, Hierarchical Reasoning Model's (HRM) parallel processing allows for what Wang (Founder and CEO of Sapient Intelligence) estimates could be a “100x speedup in task completion time.” This means lower inference latency and the ability to run powerful reasoning on edge devices. 📢 There are 3 efficiency techniques a. Single forward pass reasoning: HRM performs all reasoning inside its hidden states and emits an answer in 1 network pass, while CoT-style LLMs build a long text trace first. b. Constant-memory training: By replacing back-propagation-through-time with a 1-step gradient, training memory stays at O(1) instead of O(T), which shortens training iterations and improves GPU utilisation. c. Adaptive Computation Time (ACT): ACT version averages only about one third of the compute steps of a fixed-depth baseline yet keeps the same accuracy, so inference cost per example drops roughly 2-3X, not 100X. 🔧 Final Takeaway HRM hints that swapping endless layers for a small hierarchy plus cheap recurrence can give LLM‑level reasoning at Raspberry‑Pi costs. It also scales at inference: raise the ACT cap, and accuracy climbs further with no retraining. 🧵 Read on 👇
Rohan Paul tweet media
English
14
30
172
12.3K
Marcin W.
Marcin W.@Marcin_22NXT·
🧠 MIND-BLOWING results of 60min Yoga Nidra session inspired by @hubermanlab: ♥️ Heart rate kept ~ 50 for the ENTIRE session, not far from my 💤 RHR of 42 📊 47 min. of deep sleep detected by @whoop 🎯 NSDR - deep recovery while staying awake Tried #NSDR? Share below! 👇
Marcin W. tweet media
English
9
5
50
197.2K
Thomas Michael retweetledi
Out of Context Human Race
Out of Context Human Race@NoContextHumans·
Crab gets packaged alive and breaks out whilst at the store
English
300
174
3.7K
676.3K
Thomas Michael retweetledi
Volcaholic 🌋
Volcaholic 🌋@volcaholic1·
INCREDIBLE capture of the tsunami arriving in Kamchatka after the M8.8 earthquake struck offshore on July 30th Wow! Video via Kamchatka life
English
400
6.1K
31.3K
2.3M
🪖MilitaryNewsUA🇺🇦
🪖MilitaryNewsUA🇺🇦@front_ukrainian·
👀A huge tornado is heading towards 🇷🇺Sochi, Russia, – Russian media. 🍿
English
276
358
3K
105.9K
Massimo
Massimo@Rainmaker1973·
Incommunicability in animals
English
30
59
516
50.8K
Andrej Karpathy
Andrej Karpathy@karpathy·
2024: everyone releasing their own Chat 2025: everyone releasing their own Code
English
471
513
8K
855.9K
Massimo
Massimo@Rainmaker1973·
The side parking move by BYD cars [📹 innovology]
English
37
150
2.5K
875.6K