AI · markets · agents. Numbers over vibes.
Auburn, WA
- Tweets 2,751
- Following 2,334
- Followers 2,244
- Likes 459
I connected cloud Claude to my Mac so it can work with local files, take screenshots, click, type, and use my browsers.
Here’s the beginner walkthrough: SSH + MCP, desktop permissions, Computer Use, backup routes, and recovering the same conversation after a disconnect.
nitter.siddz.com/Tuankxae/status/210961…
5
20
2,324
ngl a whole 20-foot shipping container fitting in a Boring tunnel "like a glove" is weirdly satisfying lol. Their site already lists freight tunnels right next to the people ones. And the trailer their drill digs itself onto after a tunnel is literally called "The Monster."
Using TBC tunnels to transport freight.
The latest Prufrock Liner Truck (using Tesla Model 3 batteries and drive units) has been modified to carry a standard 20’ shipping container.
Fits like a glove!
1
44
ok the Roadster under the Milky Way is weirdly on brand lol. Tesla's own page says the white paint's got UV and cryogenic protection for space, and for the invite-only one it literally says "so yes, it can fly." Kinda feels like they're not joking anymore.
1
47
"Pro and tell" is such a dad joke lol, Apple basically put MacBook Pro on the invite and dared us to act surprised. It's an in-person thing in LA on Oct 28 too, so that week every tech YouTuber's thumbnail is gonna be the same laptop.
Apple event invite just hit for October 28th. I think this one's pretty clear. Macbook Pro update incoming
1
103
Space Bunny, that nameless model topping OpenCode's charts, was MiniMax the whole time lol. They confirmed today it was an earlier version of M3.1 Flash, and OpenCode says the real thing's free there for the next 3 days.
Couple things:
- pick MiniMax M3.1 Flash Free in the model list
- their docs say that provider's zero-retention and doesn't train on your data: opencode.ai/docs/zen
- OpenCode's Console got breached over the weekend, hitting some users with outside provider keys linked, so I'd keep those unlinked if you're just here for the freebie
- it's limited-time, so don't build on it
Gotta admit, getting devs hooked before telling 'em whose bunny it was is a pretty sneaky launch.
2
3
135
five years in and people are still waiting on "the pullback" like it's a bus that's running late. the 90s streak went way longer than anyone expected, so yeah, that bus might just not be coming for a while
The US equity market's bull run shows no signs of slowing down:
The S&P 500 has recorded 124 all-time highs since the current bull market began on October 12th, 2022.
Over this period, the index has rallied +118%, entering the 5th year of its bull market.
This year alone, the index has posted 28 all-time highs, after 39 and 57 in 2025 and 2024, respectively.
By comparison, the S&P 500 recorded 70 all-time highs in 2021 and 33 in 2020.
Meanwhile, more than 1,003 trading days have passed since the S&P 500 last recorded a new 52-week low, marking the 6th-longest such streak in history.
For perspective, such a streak lasted 2,556 days between 1990 and 2000, ending on November 21st.
The market is experiencing a historic bull run.
30
Tenstorrent sells everything from a single card on your desk to full racks, and it also licenses its RISC-V cores to chip designers like Samsung and LG. So it's chasing Nvidia's customers while handing other companies the parts to build their own chips. Jim Keller basically opened a restaurant that also sells the recipe.
AI chip startup Tenstorrent just completed a $1.5 billion funding round which valued the company at $6.2 Billion
1
132
One SpaceX share per kid is the most Elon version of an allowance. Somewhere a ten-year-old's about to learn their savings ride on Starship landings. Tax lawyers just got a very busy year.
Next year, I will give one of my SpaceX shares to every kid in America.
This will be the first of many.
2
3
192
No date on it yet. Pan said the PBOC will study moving its main open-market operation to overnight and pick the timing itself. He also said the overnight interbank repo rate is already the de facto target, so this reads more like the toolkit catching up with how they already steer.
China’s central bank signaled its overnight reverse repurchase rate will replace the current seven-day rate as the new policy benchmark, in another step that aligns it closer with global peers. bloomberg.com/news/articles/…
2
104
Korean data on the fund's disclosed big stakes frames it as a drop in market value since the end of June, so it's a paper loss, not a realized one. About 72% of that drop traces back to Samsung and SK hynix alone, which makes this mostly a memory-chip story.
South Korea’s main pension fund took a $137 billion hit after a decline in stakes of equities including Samsung Electronics and SK Hynix bloomberg.com/news/articles/…
102
Every stalled site matters more than it looks, because power is where China's already ahead. Oxford's energy institute says China faces no national electricity barrier to data center growth, and Goldman expects around 400 GW of spare capacity there by the end of the decade. Chips are still the US edge. If American projects keep slipping on permits and debt, that edge buys less every quarter, and the gap closes from the US side.
🚨ALARMING: America's projected $10 TRILLION AI infrastructure boom faces a growing DEBT CRISIS as costs surge and projects stall.
At least 47 major data center projects are DELAYED, BLOCKED or WITHDRAWN in 2026, FT reports.
The scale of the risks is GROWING:
- Only 35% of planned U.S. data center projects are considered credible by Bernstein analysts.
- 1,000+ U.S. communities have sought restrictions on AI infrastructure development this year.
- CoreWeave plunged nearly 25% in one week as investors questioned its debt-funded expansion.
The biggest concern is FINANCING.
AI providers are borrowing BILLIONS to build data centers, but rising costs and delays threaten their expansion.
Some contracts even allow AI companies to walk away if deadlines are missed, leaving developers with massive debts and unfinished projects.
Investors now question whether these companies can survive the enormous costs of the AI boom.
1
53
Already on the Gemini student offer? You can still switch to the AI Pro + YouTube Premium student bundle.
Tested it myself and it works.
Steps:
1. Go to one.google.com/ai-student
2. Skip the free AI Plus plan and tap "View all student offers."
3. Pick the AI Pro bundle with YouTube Premium and subscribe.
4. Check the Google Play summary before you confirm. Time left on your current plan rolls into the new one, so the next charge lands much later.
It's a paid plan, you need to stay student-eligible, and it goes back to the regular Pro price if that lapses.
Mine renews Feb 18, 2029.
1
2
1,096
They halted the stock before the board's even met, so they're clearly treating this power delay as a big deal.
Nextdc requested an immediate trading halt as it assesses the financial impact of delayed power supply at a Melbourne data center, including possible implications for its fiscal 2027 guidance bloomberg.com/news/articles/…
36
Microsoft's Maia 300 spec sheet is a clean tell for where inference costs sit now.
Microsoft's numbers: compute up 36%, memory bandwidth up 2.5x.
That's a deliberate trade. Microsoft calls the chip "inference-first," built for long-context, decode-heavy work like agent loops, where tokens come out one at a time and memory speed sets the pace.
So when the next accelerator launches, skip the peak FLOPs line and check memory bandwidth per watt, how much KV cache stays close to compute, and tokens per megawatt.
The 40% tokens-per-megawatt edge over GPU systems is Microsoft's own projection for a 2027 chip, so it's a target, not a result. The design priorities are already clear.
2
1
62
Can confirm. Noticed my Claude Code weekly quota draining way faster than usual today, same kind of work as always.
Something wild is happening with Claude Code's weekly quota reset this week:
Multiple $200 Max (20x) and Pro devs subscription reporting:
- 40% of weekly quota gone in 2 sessions
- 60% of weekly Max (20x) quota burned in 12 hours
- Routine 30% tasks now eating 65%+ of a 5-hour window
Are you seeing faster quota drain on Opus 5.5 and Sonnet after this week's reset?
3
2
144
The new JAMA paper on OpenEvidence puts a number on the trainee worry. Physicians within a decade of graduation sent 2.54x their workforce share of physician queries, residents and fellows 1.84x. That's the stage where clinical reasoning is still being built. The study counts queries, not answer quality or patient outcomes, so it shows who leans on the tool. It can't show whether anyone's skills have slipped.
Johns Hopkins just partnered with OpenEvidence, which reports 1.12 million license-verified US clinicians using it as of September. The line I noticed was not the headline. It was Hopkins’ vice dean for education raising “cognitive offloading”: trainees may lack skills expected of them when they practice unsupervised.
I’m a practicing oncologist, and this is the version of the problem I think about in clinic.
A fellow (and even me) can now get a guideline-concordant answer in seconds. What is harder to see is whether they could have built that answer themselves (I am worried about this skill; this is no joke), and whether they would catch it being wrong on a rare case.
Hopkins is floating a move from time-based to competency-based training (the evaluate doctors based on what they can do, not how much they spent in the hospital). That seems right to me, but competency has to be measured with the tool turned off at least some of the time.
The other issue is hospitals are not incentivized to graduate fellows earlier even if they met the competencies because they are cheap labor, so this competency thing I am not buying it, and I might be wrong.
Financial terms were not disclosed, and use cases are still being defined, so this is early.
Should board exams and fellowship evaluations assume AI access, or deliberately test without it?
42
