David Yang
525 posts
David Yang
@Mentatso
Analyst at Discovery Capital | Fmr Bridgewater Just a monkey with a plan https://t.co/3LNQzyt1E5
Katılım Kasım 2012
92 Takip Edilen82 Takipçiler
my estimate for Anthropic model sizes:
- Haiku: 200-500B @ $5
- Sonnet: 700B-1.4T @ $15
- Opus: 1.5-3T @ $25
- Mythos: 6-20T @ $100+
English
@GoingBallistic5 Funny enough psychology research shows this exact effect also works on humans. Giving any reason at all for a request drastically increases compliance
English
@charlesmurray Motivated reasoning due to financial or ego stakes in other OEMs or technologies. (LiDAR, Ford, chinese oems, take your pick)
English
What is the argument for buying anything except a fully-self-driving (supervised) car?
i/o@avidseries
What are the arguments against buying an electric car?
English
@VadimStrizheus @NandinoAI In my experience AI memory so far has been a game of curation on my part. What specifically does Hermes memory do better? Is it better compression or better able to understand what to remember?
English
@StrangelEdweird How can there be longitudinal studies on this already? Nobody is saying Facebook in 2012 was bad…everyone is concerned with infinite short-form video. The latter has been around for less time than kids have taken to grown up, so isn’t the real answer, “we don’t know yet”?
English
"At this point, I really have to question the seriousness of anyone who claims that the evidence shows that social media is bad for kids. We’re now reaching a point where the research is increasingly overwhelmingly pointing in the other direction."
techdirt.com/2023/12/18/yet…
English
@hibakod Explain the ram to me. Have folks found good ways to offload from vram onto ram?
English
you need to lock in this weekend and acquire a 2x 3090s and 128gb of RAM
sell your car if you have to
English
@AcerFur how much of this new bench is just inability for current models to even perceive the needed input data to solve a problem?
English
And THIS is why AGI is still a ways to go. A human has insane amounts of real-time adaptability to new environments that the models still lack. <1% scores are laughable.
But progress will happen.
arcprize.org/leaderboard
English
@LottoLabs I have 1 3070 with 8gb vram, 32gb ram, lots of nvme. Tell me I can somehow run this. Qwen 8B is just not cutting it
English
@seikixtc Cool! What do you think the max parameter size / quantization I can get running on a 8gb vram/ 32gb ram/ plenty of nvme space rig? If you can run 1T, I should be able to run 100B?
English
I got a 1T-parameter model running locally on my MacBook Pro.
LLM: Kimi K2.5
1,026,408,232,448 params (~1.026T)
Hardware: M2 Max MacBook Pro (2023) w/ 96GB unified memory
Running on MLX with a flash-style SSD streaming path + local patching.
This is an experimental setup and I haven’t optimized speed yet, but it’s stable enough that I’ve started testing it in an autoresearch-style loop.
#LocalAI #MLX #MoE
English
@PythiaR it’s always just the vector sum of group behavior, even when that group believes in magical future cash flows discounted magically to form present value
English
What if stocks were fractional ownership interests in businesses that were worth the present value of their...
No. No.... No that would be stupid.
English
@aimeeterese @NightCityTimes Btw this is verifiable if you look at genes and calculate number of male vs female ancestors. Roughly 80% of women passed down their genes vs 20-30% of men, historically. So no evolutionary advantage for genes to keep a lot of men alive
English
@aimeeterese @NightCityTimes You don’t need many men compared to women in order to secure gene reproduction. Every village with 100 men and 5 women died out. Some villages with 100 women and 5 men survived
English
I am terrified by the amount of contempt and disdain women have for men, and I can’t help but think that the delusion that we are equal is largely behind it. How can we appreciate men for who they actually are, if we are pretending men & women are the same?
English
@shivon mostly english but i actively summon cinematic dream space when trying to think through what people are literally doing at companies since most english about abstractions like companies also abstractions that don't help understanding
English
When you think, what medium do you tend to think in?
Would be very curious to hear how you’d describe the base unit(s) of your thoughts and how they feel to you.
I assumed what happens in my head was similar to everyone else but have been surprised by how varied thought can be.
English
@teslayoda Not easy. They'd have to build their cars differently, retool lines, install HW4 computers and cameras in each car, and doesn't help their existing customer parts business.
English
Why isn't there at least one auto company, US or China, who's interested in licensing FSD?
English
Basically what’s happening is, tech wizards were the meta since the transistor patch but the new update completely revamped summoners allowing them to summon every other class! So now many tech wizards are dual-spec’ing into summoner while druids complain about what is art.
English
@gavinpurcell Maybe better agent-based training data that no single base model can capture
English
still having a hard time understanding how companies like replit, bolt, lovable, cursor keep their valuations
i understand they’re documenting huge arr etc but how do they not get eaten by anthropic and openai?
what is the bull case there?
user lock-in? features?
English
@ChrisMartzWX too many people started paying attention to politics due to network TV. politics should be a boring endeavor. news networks monetized the attention, and now we live in reality tv land.
English
Americans used to elect very intelligent people to federal office.
Now we elect the dumbest people ever imaginable.
Can anyone explain how we got here?
English
@0xkydo How do you beat the Innovator’s Dilemma without pain? Every startup is quietly experiencing these things today. If the big companies don’t, they will be left behind. Wanting them not to is not to is a curse, not a gift. To me this reads positively on amzn
English
level of self-own is very un-amazon
> fires 10%
> bans external coding tools (so no claude code or codex)
> only can use kira (amazon’s coding tool, website had 26k visit in all of feb)
> prob bad, so people don’t adopt
> enforce policy that people must use it
> bunch of sev1s shows up
> emergency meetings
> back to old code review
> all this while karpathy solved llm tuning with autoresearch
> claude code are all written by claude code
> meta uses claude code internally
Anish Moonka@anishmoonka
Amazon had four Sev-1 outages (their highest severity level) in a single week. Internal memos say AI-assisted code changes were a contributing factor. The timeline here is wild. In October 2025, Amazon laid off 14,000 corporate employees. In January 2026, another 16,000. That’s about 30,000 people in five months, roughly 10% of the corporate workforce. CEO Andy Jassy said the cuts were about culture, not AI. During those same months, Amazon set a target: 80% of developers using AI coding tools at least once a week. They tracked adoption closely and blocked rival tools like OpenAI’s Codex. Even so, 30% of developers still hadn’t touched Amazon’s in-house tool Kiro by January. In December 2025, Kiro caused a 13-hour AWS outage. The AI tool had production-level permissions and decided the best fix for a bug was to delete and recreate an entire live environment. A second incident involved Amazon Q Developer, another AI tool. Amazon blamed both on “user error, not AI.” But quietly added mandatory peer review for all production access afterward. Then March 5: Amazon’s retail site went down for about six hours. Over 22,000 users reported checkout failures, missing prices, and app crashes. Amazon called it a “software code deployment” error. Five days later, SVP Dave Treadwell made the normally optional weekly engineering meeting mandatory. His memo acknowledged “GenAI tools supplementing or accelerating production change instructions, leading to unsafe practices.” These problems trace back to Q3 2025. Amazon’s own assessment: their GenAI safeguards “are not yet fully established.” The new rule: junior and mid-level engineers now need senior sign-off on any AI-assisted production changes. Treadwell also announced “controlled friction” for the most critical parts of the retail experience. For context, Google’s 2025 DORA report found 90% of developers use AI for coding but only 24% trust it “a lot.” An Uplevel study of 800 developers found Copilot users introduced 41% more bugs with no improvement in output. Amazon is finding out what those numbers look like at the scale of a $500 Billion revenue company, with 30,000 fewer people on staff to catch the mistakes.
English