Michael Laskey

233 posts

Michael Laskey banner
Michael Laskey

Michael Laskey

@Michaellaskey7

shipping robots CTO @sheeprobotics | ex @berkeley_ai Phd

Katılım Ağustos 2019
126 Takip Edilen1.2K Takipçiler
Sabitlenmiş Tweet
Michael Laskey
Michael Laskey@Michaellaskey7·
When I Joined @sheeprobotics , I saw a company with an insane vision laid out by @MurtyNag . Billions of low-cost robots sustainably caring for our planet. I am very excited to announce my new title as Chief Technology Officer to make this vision a reality.
Michael Laskey tweet media
English
12
1
90
13.2K
Michael Laskey
Michael Laskey@Michaellaskey7·
if a sub <100 mm company can build deepseek no one leads ai right now and there is alpha in being under-valued
English
0
0
3
246
Michael Laskey
Michael Laskey@Michaellaskey7·
my partner switched to deepseek cause the whale was cute .... b2c is rough
English
0
0
0
234
Michael Laskey
Michael Laskey@Michaellaskey7·
@DJiafei robotics with sim needs really low resources to work ... we train and ship general robots with <100k compute
English
0
0
2
119
Jiafei Duan
Jiafei Duan@DJiafei·
Deep-seek sent shockwaves through the industry on the eve of Chinese New Year, causing NVIDIA to drop 16.97% (over $500 billion). It got me thinking: perhaps more isn’t always better. Compute may not be the answer to everything. Could this idea also apply to robotics foundation models and the many well-funded robotics data companies out there? What might be the ‘deep-seek moment’ for robotics that challenges our assumptions about data and scale
English
3
0
16
2.1K
Michael Laskey
Michael Laskey@Michaellaskey7·
Lol ... yeah you can't stop this technology. Its just not that hard to build this stuff is the core issue. Unlike nuclear or missiles, this is pure software innovations.
Alexandr Wang@alexandr_wang

DeepSeek is a wake up call for America, but it doesn’t change the strategy: - USA must out-innovate &race faster, as we have done in the entire history of AI - Tighten export controls on chips so that we can maintain future leads Every major breakthrough in AI has been American

English
0
1
8
1.9K
Michael Laskey
Michael Laskey@Michaellaskey7·
id argue CoT style RL is more like MPC then model-free RL
English
0
0
1
299
Simon Kalouche
Simon Kalouche@simonkalouche·
You already have specialized machines for everything and that won't change. You have a coffee maker to make coffee, a vacuum cleaner to pick up crumbs, a lawn mower with blades to cut grass, a car that drives on roads, washing machines that wash our clothes, dish washers to wash our dishes and the list goes on. In the next couple years every machine will evolve into a robotic version of itself. The world will not evolve backwards to having humanoids hand scrubbing dishes, hand washing clothes, hand trimming the lawn with scissors, driving manual cars, picking up crumbs with their fingers, etc and all the companies that make vacuums, dishwashers, lawn mowers arent going to sit around and stop innovating with new products just to give humanoids a purpose. The world will just not evolve this way. We can all want humanoids (as do I because they're clearly cool) but that doesn't mean they will ever make sense commercially and for them to exist at scale and last they have to make sense and be commercially viable. In 10 years when humanoids can actually do real tasks there will have been a Cambrian explosion of new multi-purpose robots that will leave humanoids searching for purpose more than humans do.
English
2
0
6
343
lala
lala@zxlava·
if you know your task ahead of time (run, fly ...) then yes biomimicry is dumb - you should optimize solely for that specific task but there's likely a strong market for human-compatible machines that can handle multiple tasks, even if they perform each one less efficiently it's still science fiction and a LONG way out, but many people (myself included) would prefer one end-to-end humanoid rather than like 50 new specialized machines in their home i don't really need a task-optimized robot just for dog walking, even if it'd outperform a humanoid. a human compatible form will be good enough for me! this suggests 2 distinct markets will emerge - performance-focused customers who'll want specialized robots optimized for specific tasks - convenience-oriented customers who prefer a single humanoid robot that can handle multiple tasks adequately, even if not optimally
Simon Kalouche@simonkalouche

Feynman on why humanoids and biomimicry are silly.

English
2
0
7
995
Michael Laskey retweetledi
Jiafei Duan
Jiafei Duan@DJiafei·
Every time I watch this video, I can't help but wonder: why don't we have robot butlers in our homes yet? The hardware seemed capable 14 years ago with teleoperation. Is it just the "robot brain" we're missing, or is there more to the puzzle?
English
37
43
336
47.4K
Michael Laskey
Michael Laskey@Michaellaskey7·
if o3 is using a task specific process reward model that is learned from supervised data. Isn't that the definition of narrow ai?
English
0
0
0
173
Michael Laskey
Michael Laskey@Michaellaskey7·
@fchollet is it really a break through though if you collect a bunch of training data for the process reward model? I feel like it still is task specific narrow ai; in that regards.
English
0
0
1
90
François Chollet
François Chollet@fchollet·
People scaled LLMs by ~10,000x from 2019 to 2024, and their scores on ARC stayed near 0 (e.g. GPT-4o at ~5%). Meanwhile a very crude program search approach could score >20% with hardly any compute. Then OpenAI started adding test-time CoT search. ARC scores immediately shot up.
English
53
187
2.4K
185.1K
Michael Laskey
Michael Laskey@Michaellaskey7·
Its wild the first e2e self-driving car was in 1989 Makes you really appreciate how hard it is to scale robots.
GIF
English
1
0
6
374
Pranay Thangeda
Pranay Thangeda@pthangeda_·
@Michaellaskey7 How do you even go about getting insurance in these scenarios? Do traditional commerical liability policies cover it or do they have clauses excluding autonomous equipment operation?
English
1
0
0
89
Michael Laskey
Michael Laskey@Michaellaskey7·
Try selling an agent controlling commercial grade lawn mowers .... whole different level of liability challenges
BURKOV@burkov

When considering the acquisition of AI agent solutions or engaging contractors to implement agents in your workflows, ensure your contract includes an indemnification clause holding the provider liable for any material or financial losses directly attributable to the agents' actions. Then see how many providers will be willing to sign.

English
2
1
6
2.4K
Michael Laskey
Michael Laskey@Michaellaskey7·
i think the biggest lesson for anyone looking to ship robots is to really appreciate the difference between 90% and 99.9% that is two orders of magnitude in reliability and yes your customer will expect it
English
8
5
126
7.2K
Michael Laskey retweetledi
chris
chris@cwamidon·
Hardware founders: stop being cowards. I know what you say in private. Stop bowing to VCs because you want your next round. Speak your actual mind. Venture returns have been shit for years. They can’t afford to blackball good companies because you disagree with them.
English
12
6
289
39.2K
Michael Laskey
Michael Laskey@Michaellaskey7·
Great master plan ... you learn so much by shipping regardless of form factor.
vivek 🦀 🤖@jokrvivek

Robotics too @maticrobots master plan: 1. Make the best robotic vacuum (Market: 20 mil sold in 2024; growing @ 30% YoY📈) 2. Generate free cash flow 3. Use ☝️ money & all the learnings to invent *and ship* a home declutter robot 4. General purpose humanoid We've started step 3

English
0
0
4
829
Baran Cezayirli
Baran Cezayirli@barancezayirli·
@rohanpaul_ai 25% of the code you write is boilerplate code. Most senior devs in large teams contribute 10 - 15% of the codebase. It is not the lines you write but how you solve the problem with a simple, maintainable, scalable approach.
English
6
2
155
12.9K
Rohan Paul
Rohan Paul@rohanpaul_ai·
Connecting the dots 😃 Google laid off its entire Python team in April 2024 as part of a cost-cutting effort. And today it's revealed on the earnings call that more than 25% of all new code at Google is generated by AI. Software Engineering will disrupted in more ways by AI than we can see right now.
Rohan Paul tweet media
Rohan Paul@rohanpaul_ai

Let that sink in. "more than 25% of all new code at Google is now generated by AI." I think that the next 10 years is the last period that any human will ever write any code.

English
116
176
1.1K
322.6K
Michael Laskey
Michael Laskey@Michaellaskey7·
Got to give a talk on what we are building at @sheeprobotics . Agents for outdoor labor are coming.
Electric Sheep@sheeprobotics

Because of the lack of real world data feedback loops - the chatgpt moment for robotics isnt going to happen in in the lab. We need to build deployment engines at the scale of a Tesla - bootstrapped by sim-trained models. Hear our CTO @Michaellaskey7 talk about how we are building and scaling end to end ML for outdoor work.

English
0
0
5
366

Keşfet

@DJiafei @simonkalouche @zxlava @fchollet @pthangeda_ @sheeprobotics @elonmusk @BarackObama