Their claim is the boom won't actually reach ground level for overland flights due to how they're profiled.
Of course that's the theory. The Trump Admin just allowed for a fairly audible boom.
4,230 karma · joined March 24, 2011
CTO @ stealth AI related startup
Of course that's the theory. The Trump Admin just allowed for a fairly audible boom.
"My personal impression is that within these quantizations Qwen 3.6 27B is as good as (or maybe slightly better than) DwarfStar4. Though, I won’t be surprised if for longer context projects DS4 has an edge."
I feel like the amount of context bloat that OpenCode puts these small models into the dumb zone too quickly. The system prompt alone is 9k tokens, and when you add your own setup it can easily creep up to 15k.
They have a newer model that is slightly slimmer w/ more features like active noise cancellation, but has a smaller battery that is marginal for full-night sleeping, so I didn't bother upgrading.
One thing to note is they have some bad reviews due to issues with batteries dying or depleting quickly - my right bud wouldn't charge past 50% after 9 months of daily usage, which only lasted about 6 hours or so before dying. That said, Anker's tech support promptly sent out a new pair with minimal friction, which I received just days after reaching out. I've had 4 months of usage on those new ones without a hitch. Hopefully those kinks have been worked out.
Even if these new ones die outside of warranty, they're so good that I feel I'll have gotten my money's worth and will immediately buy them again.
Also a simple noise machine can work wonders.
He discussed this topic and how generally it's left to those who are more notable in a field to ask the 'dumb' questions everyone else is afraid to ask. And such questions often need to be asked to get the audience on board and open the floodgates with areas of niche research - the speaker themself is often too far into the rabbit hole to discern the difference between opaque and obvious.
So it stands to reason, at smaller conferences this would be a big problem, with fewer thought leaders in attendance whose reputations are intact enough that they wouldn't mind looking foolish.
You're writing words that I did not say or imply.
The point is going against any (current) admin is almost always bad for a publicly traded company. Any public entity is going to have to have extremely good reasons to "fight back", how doing so is good for business. As a CEO of such an entity you're having to answer to many people who want a concrete plan and a belief in your strategy.
In the first rodeo, when all this was novel, it was believed such social signaling would pay off. Obviously silicon valley as a whole no longer feels this way.
TSLA is an outlier being grounded more on some superior man theory, that Apple did have in the past w/ Jobs, who is no longer there. Religious fervor stuff. It doesn't really apply. Rational moves here, please.
> There are myriad other things he could’ve done, that have a strong argument towards higher shareholder value
This is what I asked you to expound on. Please state a few.
That said, I'd love to enlightened to how it's myopic, or rather, what course(s) of action you would take, keeping in mind that Apple is a multi-trillion dollar public company.
It would be unusual for a leader of such a thing not act in accordance w/ shareholders' best interests, as well to defy likely board guidance.
Just start small, record 4 bars of something, play it back. Use a metronome to gauge the actual amount of drift. See if you can hear what the metronome is telling you. Try to improve it with a few more attempts, then call it a day.
Try again after a good night's rest. Over time you should be able to naturally feel when you're slowing down or speeding up.
She stressed how important it was to record myself and listen back, in fact she encouraged me to do it for hours at a time. The immediate reaction the first few days of trying this was "holy crap, not only am I pitchy and can't hear it naturally, I'm constantly slowing down and speeding up like +/- 10bpm."
The experience was so distressing that I tried to quit my next lesson, but she pushed back with "Hey if you can hear it, you can fix it. It won't be tomorrow, or next week, or even next month. It could take a year. Improvement happens little by little. And I guarantee you'll see progress as it happens. But you have to put in the work."
After a few weeks of working up to it, I settled into a pattern of spending ~3-5 hours most weeknights in the darkness of my closet, recording myself playing Beatles songs along with an acoustic guitar into a 4-track. Usually just going back and forth over 4-8 bars of a song for 30 minutes, then 30 minutes another section, really just focusing on a couple songs like that. Toward the end of the session I'd attempt several full run throughs, get super frustrated (over increasingly minor issues), and end the session.
And she was right, by about the third month I was comfortable enough to perform in front of the person I was seeing, and by month five, I could get through a song with barely any mistakes, maybe one out of three chances. By the ninth month, after a 15 minute warmup, I could get through a 3 song stretch with just minor errors, enough not to totally embarrass myself at an open mic night.
At that point I felt I hit my goal and took a break from lessons. Never did an open mic night. Continued practicing a bit in my closest, but after a month or two I stopped as well.
And here 20 years later, my rhythm actually is pretty solid... I've been a consistent bedroom guitarist, and routinely record myself, and sometimes I don't bother with a metronome because it sounds that consistent. That said, I stopped singing and that ability is completely gone. But I am starting a similar process learning classical guitar.
So I go back to that original bit of advice with just about anything I try to tackle now... if you're self-aware of your issues, and can actually critically hear them in a recording, then there is a path forward.
- Yes food, as well as alcohol, was quite cheap. Had very few meals that came out to more than $10, alcohol (about $3-4/drink) included.
- I purchased a couple pairs of running shoes that were about 30% cheaper than they were offered for sale in the US.
- I purchased an umbrella for $45 that sells in the US for $75.
- An all-access pass at their premier amusement park, Fuji-Q Highland, was only about $40 - when entry to comparable parks in the US can easily be twice as much.
- I recall the subway came out to around $1.50 a ride, roughly half what the NYC subway costs and the 1 and 3 day passes made it ridiculously cheap (IIRC something like $5/$10).
- I only used capsule hotels, but those were only $15 to up to $38 for a luxury one, almost all in desirable/touristy areas.
- I also took a look at apartments, and in decent areas in Tokyo you can find small apartments for about $1500 that would cost ~$3500 in Manhattan, or maybe $2000 in medium sized US city centers.
That was bargain basement for that era. IBMs, Compaqs and the like were ~$5k similarly configured, and the first 486s were in the $7-9k area.
I guess what might be more interesting is how many people bet $1000 the day before that, and the day before that. That would be more helpful to determine what is noise from well-informed outsiders vs. insiders.
I remember beating Zaxxon when I was only 6 years old in '82, in large part because I probably spent at least 100 hours watching older kids do it.
Claude OTOH works better with ambiguity, but it also tends to stray a bit off spec in subtle ways. I always had to take more corrective action w/ the PRs it produced.
That said, I haven't used CC in 3 months and the latest models may be better.
- I don't update plans in the past - any work that superseeds work from an earlier plan is simply a new plan. If during creation of a new plan I review the plan and decide I want to something else that requires a spec update, I trash the plan, do the spec update, and rerun plan generation. Past plans of course can point to divergent specs but that's not something I care about much, as plans are a self-contained enough story of the work that was done.
- Specs: these are generally static, but updatable as the project evolves. And they're broken out to an index file that gives a project overview, a high-level arch file, and files for all the main modules. Roughly ~1k lines of spec for 10k lines of code, and try to limit any particular spec file to 300 lines. I'm intimately familiar with every single line in these.
- Plans: these are the output of a planning session with an LLM. They point to the associated specs. These tend to be 100-300 lines and 3 to 5 phases.
- Working memory files: I use both a status.md (3-5 items per phase roughly 30 lines overall), which points to a latest plan, and a project_status (100-200 lines), which tracks the current state of the project and is instructed to compact past efforts to keep it lean)
- A planner skill I use w/ Gemini Pro to generate new plans. It essentially explains the specs/plans dichotomy, the role of the status files, and to review everything in the pertinent areas of code and give me a handful of high-level next set of features to address based on shortfalls in the specs or things noted in the project_status file. Based on what it presents, I select a feature or improvement to generate. Then it proceeds to generate a plan, updates a clean status.md that points to the plan, and adjusts project_status based on the state of the prior completed plan.
- An implementer skill in Codex that goes to town on a plan file. It's fairly simple, it just looks at status.md, which points to the plan, and of course the plan points to the relevant specs so it loads up context pretty efficiently.
I've tried the two main spec generation libraries, which were way overblown, and then I gave superpowers a shot... which was fine, but still too much. The above is all homegrown, and I've had much better success because it keeps the context lean and focused.
And I'm only on the $20 plans for Codex/Gemini vs. spending $100/month on CC for half year prior and move quicker w/ no stall outs due to token consumption, which was regularly happening w/ CC by the 5th day. Codex rarely dips below 70% available context when it puts up a PR after an execution run. Roughly 4/5 PRs are without issue, which is flipped against what I experienced with CC and only using planning mode.
What's interesting is the one in use today - from the early 17th century - is not the most modern variant. There was another revision from the mid-19th century that fell out of favor because it sounded a bit off, less rhythmic, less sacred (ie. Kingdom -> Government).
I started my CS degree in '94, and my junior year was... weird. Suddenly a quarter of the faces in the lab were not the types you would ever expect to see there.
The first half has me thinking instant classic, my hope is sky high. But then toward the end I find myself looking at my watch and realize it's simply not going to the stick the landing.
OTOH, many acclaimed streaming series have generally done this well. My take is that as long-form storytelling has evolved, movies have transitioned into this post-modernist phase as directors/writers don't feel they have the runway to tell something truly cohesive that doesn't end up being trite. It's much more about saying 'something' or imbuing a feeling than telling a fully fleshed 3 act story.
Cyberpunk has changed my perspective on this. Started last August, now 84 hours in, and I plan to do a completion-ist run of at least 120+ hours. Around hour 50 I could have completed the main storyline which snuck up on me quicker than expected, so I'm trickling in the Phantom Liberty quest-line while chewing thru gigs and side jobs, many of which add a ton of narrative impact.
I remember way back when Sopranos ended and I was fairly sullen for a few months after... similar response to The Wire, Mad Men, Breaking Bad, BCS, Succession. Don't watch TV as a rule, but I immediately stopped what I was doing when a new ep of those dropped. Felt almost like I lost a loved one when each ended.
Pretty sure I'll have the same response to this game, I'm dreading leaving this universe. After that I'll get to Baldur's Gate 3 and probably have the same experience. Then after taking a year to finish that off, will likely move on to RDR2. Maybe by then I'll be able to play GTA6 on my Mac.
Basically, my thinking is that - a premium series is going to be roughly 50-60 hours, so if the story and immersion is there... which it certainly is w/ a game like Cyberpunk, the base game should roughly give a similar playtime. The fact that there is so much bonus content that is lovingly crafted gives players the option to stay longer if they so desire is fantastic given these things are now taking 10+ years to produce.