It's not clear that writing poetry is a bad use case. Reasoning models seem to actually do pretty well with creative writing and poetry. Deepseek's R1, for example, has much better poem structure than the underlying V3, and writers are saying R1 was the first model where they actually felt like it was a useful writing companion. R1 seems to think at length about word choice, correcting structure, pentameter, and so on.
Ok, that makes some sense. I guess I was thinking more about the creative and abstract nature of poetry, the free flowing kind, not so much about rigid structures of meter and rhyme.
Indeed. I would assume that a reasoning model would do far better at things like actually maintaining meter or rhyme scheme, something that models (even with good attention mechanisms) generally do very poorly at.
To be blunt, an AI isn't a good tool for writing poetry either. At least, not the kind people read as a high literature form. For commercials, jingles, Hallmark cards, etc. sure
There exists poetry that requires a lot of mathematical understanding! This is "literally" (and I mean literally in the literary sense) from a Stanislaw Lem story about an electronic bard, translated from Polish by Michael Kandel:
A love poem, lyrical, pastoral, and expressed in the language of pure mathematics. Tensor algebra mainly, with a little topology and higher calculus, if need be. But with feeling, you understand, and in the cybernetic spirit.
Response:
Come, let us hasten to a higher plane,
Where dyads tread the fairy fields of Venn,
Their indices bedecked from one to n,
Commingled in an endless Markov chain!
Come, every frustum longs to be a cone,
And every vector dreams of matrices.
Hark to the gentle gradient of the breeze:
It whispers of a more ergodic zone.
In Riemann, Hilbert or in Banach space
Let superscripts and subscripts go their ways.
Our asymptotes no longer out of phase,
We shall encounter, counting, face to face.
I'll grant thee random access to my heart,
Thou'lt tell me all the constants of thy love;
And so we two shall all love's lemmas prove,
And in our bound partition never part.
For what did Cauchy know, or Christoffel,
Or Fourier, or any Boole or Euler,
Wielding their compasses, their pens and rulers,
Of thy supernal sinusoidal spell?
Cancel me not -- for what then shall remain?
Abscissas, some mantissas, modules, modes,
A root or two, a torus and a node:
The inverse of my verse, a null domain.
Ellipse of bliss, converse, O lips divine!
The product of our scalars is defined!
Cyberiad draws nigh, and the skew mind
cuts capers like a happy haversine.
I see the eigenvalue in thine eye,
I hear the tender tensor in thy sigh.
Bernoulli would have been content to die,
Had he but known such a squared cosine 2 phi!
From The Cyberiad, by Stanislaw Lem.
Translated from Polish by Michael Kandel.
Here's a previous discussion of Marcin Wichary's translation of one of Lem's stories from Polish to English. He created the Lem Google Doodle, and he stalked and met Stanislaw Lem when he was a boy. Plus a discussion of Michael Kandel's translation of the poetry of the Electric Bard from The First Sally of Cyberiad, comparing it to machine translation:
>Lem’s fiction is filled with haunting, prescient landscapes. In these reissued and newly issued translations — some by the pitch-perfect Lem-o-phile, Michael Kandel — each sentence is as hard, gleaming and unpredictable as the next marvelous invention or plot twist. It’s hard to keep up with Lem’s hyper-drive of an imagination but always fun to try.
Similarly, if you've had LASIK and struggle with halos and starbursts around taillights and headlights, try increasing the brightness of your instrument panel. It will cause your irises to close in response to the brighter light, which allows less light in through the uncorrected area of your corneas. Basically: in LASIK surgery they only apply correction to the very center area of your cornea, and the more your irises open, the more of those higher order aberrations you will see.
Interesting. I had "S.M.I.L.E" surgery about 8 years ago, and I find the opposite to be better for me. At night, I dim my instrument panel to as dark as possible.
Text on the homepage hero: "Boost Revenue and Reduce Churn with AI-Powered Dunning"
Today I learned the word "dunning" -- I'd never heard of it before. Except as the name of one of the researchers after whom the Dunning-Kruger effect was named. Per Wikipedia: "a cognitive bias in which people with limited competence in a particular domain overestimate their abilities." -- So, due to pattern-matching, the word "Dunning" immediately lit up the same neural network in my psyche that activates when I think of over-confident idiots.
Might want to avoid using a word that evokes over-confident idiots when selling your AI-based subscription billing recovery solution.
I thought that at first, but then realized that it is, in fact, correctly a client error: the client erroneously directed a coffee-related request to a server that is, in fact, not meant to service such requests. :-P
They have to handle it anyway. There's nothing in the HTTP specifications that disallows a server from using whatever codes it wants to that aren't specifically specified. There is no "HTTP 527 Server Needs a Nap" but it is perfectly legal for my server to reply to clients with that HTTP status, and clients are expected to handle it like they would handle any other non-specific error in the 5xx class of response codes (server error).
Other perfectly legal responses:
242 - TOO MUCH COFFEE
(Server is overcaffeinated and processing requests too quickly)
299 - SUCCESSFUL BUT SASSY
(Request succeeded but the server is throwing shade about it)
333 - QUANTUM UNCERTAINTY
(The server simultaneously succeeded and failed until observed)
Oh, definitely hard disagree with this! I am an engineer because I love it and because it brings me joy, and I love love LOVE things that involve humor and fun within the realm of what is oftentimes just work. It delights me when people can find ways to be creative and tongue-in-cheek and not take things so seriously all the time. It is one way in which we can have fun in our work. For me, the idea of keeping all my fun separated from my work sounds like a dystopian nightmare!
"pay to feed in the surplus"? In my U.S. context, this sounds surprising: are you saying that sometimes the energy utility CHARGES you money when you feed solar energy to the grid? That sounds... bizarre?
If I were to deliver you 1000 gallons of milk that expires in 10 minutes*, would you pay me or want to charge me for that priviledge. Assume you have enough milk already.
* (< 1 millisecond for electricity but hey for the milk analogy say 10 minutes).
I don't think this is an apt analogy. Electricity is pulled, not pushed. (Well, I realize it is actually a bit more complicated than that, but for the purposes of this conversation, I think the generalization is sufficient.)
If there is no grid-side demand for the energy from my solar panels, then they're not going to be force-feeding it to a grid that doesn't want it.
When you plug a light bulb that's rated to draw 20 watts into a power source capable of delivering 1000 watts, it doesn't blow up your light bulb, because the light bulb only takes what it needs from the source.
It is more complicated on grid level. Some inverters might be smart enough, but some are not. Thus there is both push and pull going on. And this can affect voltage and frequency.
This is the only reply I understand, so I’ll follow up here - in this case, is the end user paying to push the energy into batteries that the power company owns? Or is the extra energy just dissipated somewhere? Or does the power company stop producing their own energy once the grid is saturated? (I believe that electricity doesn’t actually flow like water, so it’s not the exact same electrons that the solar panel is producing which would be stored on batteries or dissipated somewhere)
[edit] if you don't produce the exact amount consumed at the exact same time, the current will loose its phase and the effective energy transported will be reduced, while producing surtensions at different grid points (don't quote me on that, but accumulator try to keep the phase, and too much divergence can cause their failure from what I understand. But I'm no physic major and I might be completely wrong). Grid coordination is vital.
The main issue with consumer solar connected to the grid is the lack of predictability. The reason why we charge if you put electricity on the grid when too much energy is produced is because we did not plan on offsetting this production, and the network people have to keep it balanced. Reducing the output of thermal fossil plants is usually free, but while we can module nuke plants, it's best and way cheaper when that is predicted (also, opportunity cost is high, so for privately owned nuke plants, it cost even more. EDF is the bitch of the European electricity market so they eat the opportunity costs, but they are the only plant owner who does it (and I won't talk about ARENH here, but again, EDF is the bitch of the EU).
Then, when really to much shit is on the grid, and you have reservoir space, you pump the water up the STEP (batteries, but better). Up to 4500€/mwh if unpredicted (it's never actually that, but it's the price seer in the SPOT command order). And lastly, if it's the only solution, wind/solar farms are shut down. Again, if predicted, good time to do the maintenance.
In the future consumer grade installations will probably come with a lot of stuff to help with the lack of predictability. I don't love it, but that's actually my current job (well, I actually love the actual job, really interesting shit, I'm not so sure about the moral implications of even more surveillance. Because we do have the geoloc of the newer installations. We don't link it with anything relevant yet, and it's anonymous for the moment, but will it stay that way?)
Is there a reason that consumer solar can’t be accounted for as smaller stable nodes in the system (sounds like they are still considered volatile and not a reliable source of inflow power)? For instance, maybe a rooftop solar array can easily be sucked up if the home decides to do a few loads of laundry that day, and therefore since there aren’t enough homes with solar arrays, it’s harder to predict an average influx per day from consumer solar to power the grid?
Are there any goals to shut down larger plants or not build larger solar fields by instead subsidizing distributed solar on peoples’ property?
Utilities are so interesting. The other afternoon I was looking out at the hills as the trees change, and said to my friend, “what a beautiful view… besides all these power lines! Although, I’d rather have the infrastructure than an unmolested view…”
> Is there a reason that consumer solar can’t be accounted for as smaller stable nodes in the system
It's not my subject (i mostly work on automation for the network and security teams), but when i consider some changes we did last year, i think the issue with consumer-grade solar is that we didn't know where installation number XXXX was set, not even which country it was from. That is changing, we will now know if it's near Paris, near Munich, near Barcelona... That will probably help with output prediction. I think at some point there were talk about using geoIP, but that was shut down for some reason (i think it was a mix of geoIP lying, and privacy/GDPR considerations we weren't ready to tackle on yet).
When there's too much energy available for the grid, and the price goes negative, producers are paying for someone to use that energy or paying another potential producer to reduce their output.
Some industrial users have variable demand, and a lower (or negative) price could encourage them to use more. A multi-region internet service might send more traffic to a datacenter with negative electricity prices, even if in increases latency for users.
Some producers need time to modulate output, and stopping and restarting can be expensive. Solar and Wind are at least technically easy to start/stop, but subsidies may make it economic to pay the grid to deliver electricity; either because of contracts/subsidies, or because the expense to deliver unwanted electricity is less than the expense to monitor pricing and reduce production.
What happens if no one takes the power (not one solar installation but lets assume lots of surplus power). Does it screw up the grid? Increase voltage/frequency?
Yes, if nobody removes the excess, grid frequency will increase. Running too high or too low frequency can damange equipment, on both the supply and demand sides.
When there is a lot of wind and sun simultaneously there is effectively too much electricity supply on the net, and day-ahead prices drop negative. This is partially fueled by flat subsidies per unit produced (has been fixed for a while for new installations), so producing when there's oversupply can still be profitable. Also most households have a flat energy rate and can amortize their energy usage over the year, so they will always keep their solar panels on even if the energy is less than worthless.
US pricing works very differently, especially in Texas.
Purchasers want to make margin on anything. So they charge this on the seller and then on buyer. Also with how markets work it is not like there isn't risks or need to at least some level forecast how much you will be buying and then selling. So their cut also covers this work.
And finally electricity is a spot market. So there is agreed price for certain period and in some cases market can be distorted and that price can be negative. It could be idiotic subsidies or production that can't be ramped as effectively. Or someone does massive mistakes with their bids, think of trading bots going wrong.
Genuinely, I'm curious to hear from OP if you explored using filesystems that can provide de-duplication, snapshots, versioning, and compression and compared that to the SQLite approach. It would be interesting to get your take on the comparative benefits of SQLite compared with an approach that uses a more advanced filesystem like ZFS or btrfs. Once you have a filesystem that can de-dup, atomic changes are more or less a matter of building up a directory with the new version first and then swapping one directory for another with a rename. Though I assume there may be a small race condition in between the two directory rename calls. I don't know if there's a standard call to tell the filesystem "mv current old && mv new current" in one shot.
Clace is meant to be portable and easy to operationalize. It runs on Linux/macOS/Windows. The only external dependency is Docker/Podman for starting containers. Depending on particular filesystem features would make it much more difficult to setup, I did not explore that. There would be use cases where using filesystem features would be more appropriate.
If you are using Docker, then you know the target system is Linux (unless you are building windows containers). There is no reason not to rely on filesystem features.
Clace runs natively on Windows and macOS, it does not have to run in a container. It can then start containers, either Linux or Windows (the specs https://clace.io/docs/container/overview/#app-specs are Linux by default)
reply