- “(The) honest caveat:” (or “genuine caveat:”, both with the colon)
- “(The) honest answer:” (again, with colon)
- “The thing to internalize:”
- “The smoking gun:”
(really, sentences that start with “The <tag suggesting the next clause is the key point>:” are a strong tell, but those four are the most prolific)
- “load bearing” (when not talking about architecture)
- “blast radius” (when not talking about actual explosives, but rather the effect of an event/action)
- “smoke test” (esp. when “sanity check” is more apropos)
- Lists of three clauses/adjectives where the third is really just a combination of the first two
- Referring to the “shape” of things figuratively
- Social media posts that end with “Curious if anyone…”
- Stories or anecdotes using. “Oh. Oh.” (where the second “oh” is italicized)
Edit: Yes, some of those last ones are terms that we often use as devs...but I would argue about the actual frequency of their use. Plus, these tells live on in prose generated by the latest models.
It’s just still so trivial to jailbreak even the latest Anthropic models (via api, and not talking about the silly ENI or Pliny breaks) I don’t understand where the safety teams are doing their work. Is it in the default chat-trained model?
CarPlay is essentially a conditional pair of video inputs. Any system that supports on-screen rear-view camera and that has a wheel speed sensor can support CarPlay.
Well done! I’ve built this same sort of thing for my family to play with. My advice for the best results:
1) Structure the choices offered by the LLM; add “choice_type” and add instructions to the LLM on what those choices should do. E.g. action, dialogue, investigation, whatever makes sense for the genre—the LLM can even generate these at story start—then “choice should direct the narrator to focus on an action-oriented moment”, “choice should direct the narrator to focus on a dialogue between two or more characters in the scene”, etc.
2) Use reasoning whenever making tool calls for choices, summarize the reasoning, and include it in narrative summaries provided as part of the context for future narrative requests. For example, the combined summary might be: “In the last narrative I wrote for the user, Harry and Luna were startled by the noise coming from the edge of the forest. Important scene developments: 1) Luna and Harry had been approaching the edge of the forbidden forest for the last three narrative turns, and in the turn I just wrote they arrived at the edge. 2) Harry seemed to be the more courageous of the two in previous narrative turns, but in the most recent one, the user’s choice resulted in Harry becoming more deferential to Luna. 3) In the most recent narrative turn, the noise that had been emanating from the forest was now accompanied by a flickering light. I then suggested paths that would allow for character development through dialogue between Harry and Luna (I gave two options here), a path to move the story forward by having Harry take Luna’s hand before running into the forest, and another path that would slow the pace by having Luna investigate the flickering light accompanying the sound. The user’s choice: ‘Have Luna investigate the flickering light.’
3) Add an RNG weighted by story length or whatever works for you that will result in choices that lead the story to a conclusion. Include that direction in the tool call for generating choices, along with a countdown to the finale choice.
This is a rough mental sketch of what worked the best for me, i purposefully left out implementation or application details, as I don’t know what you’re wanting to do next.
My answer to this in my own pet project is to mask terms found by the NER pipeline from being corrected, replacing them with their entity type as a special token (e.g. [male person] or [commercial entity]). That alone dramatically improved grammar/spelling correction, especially because the grammatical "gist" of those masked words is preserved in the text presented to the LLM for "correction".
Exercise vigilance regarding copycat or coat-tailing sites that seek to exploit the project's popularity for potentially malicious purposes. It is imperative to rely solely on information from https://Helper-Scripts.com/ or https://tteck.github.io/Proxmox/ for accurate and trustworthy content.
One note: For truly "responsive" text (and other measures, like padding/margins), I often use relative units of measurement. At the very least, rem/em, but more and more I'm using viewport¹ (including dynamic) and container query² units. I'm not your target market, and I know that owning the renderer makes this request more complex that it would seem, but I thought I'd point it out just in case you think it should be on your radar.
Well, my wife and I have been on a months-long experiment. We have HomePod minis, Macs and iPhones/iPads in the house. They are able to access the internet without restriction (other than using my own DNS resolver for ad/malware blocking purposes.
Our TVs (2020-era Vizio and 2018-era Samsung) are on a separate VLAN for home automation control, and are otherwise blocked from the internet¹. Additionally, they have the various "content intelligence" features disabled...just in case.
We also have a few Nest devices (the 1st gen wired Hello doorbell cam, The Nest/Yale deadbolt, a 2nd gen thermostat, and some Nest Protects) that are normally similarly segmented, though the Hello is allowed to communicate to the necessary domains for video streaming and PubSub notifications.
On August 1, while on a neighborhood walk without any electronic devices, we formulated the plan: every day, we'd find a reason to discuss mulch² in the presence of various devices in our home. What color of mulch we think would look best around various trees. The virtues of recycled rubber as a mulch substitute. The drainage issues it causes. And so on.
We committed to never searching for mulch online (to hide from the ever-present surveillance online), never discussing it with anyone (to avoid social network effects), never buying it (no data broker can hoover up mulch purchases), not dwelling on any social media post about mulch (analytics, man, it's crazy what that bit of metadata can do)...not even hanging around the garden department of local stores (gotta avoid bluetooth/BLE/wifi tracking).
But I DID disable the DNS blocklists (much to our browsing frustration). And while the smart home stuff remained on its own VLAN, I allowed it otherwise unfettered access to the internet during the month of August.
Since the experiment began, we've seen the net sum of zero (0) targeted ads about mulch. No banners, no interstitial social media posts, no phone calls, no flyers in the mailbox. Nothing.
I really don't believe that our devices are eavesdropping on us, but in the interest of science, the experiment continues for another month.³
---
1) Yes, I recognize that Sidewalk/ethernet-over-HDMI/hard-coded DNS/etc is a purported "thing", but I don't believe it's likely. I'm controlling for this during the month of September by re-enabling the filtering mentioned at the start; if our TVs are committed to exfiltrating surveillance data.
2) We've not really been discussing mulch. I'm using that as a proxy here, because all of the internet is a series of tubes that lead to advertising networks. But we did choose a unique topic of conversation that would be relevant to our demographics, geographical location, and season, and meaningful to advertisers.
3) On September 1, I re-enabled all the blocklists and VLAN network filters/blackholes. But we continue to discuss, er, mulch. Like I said, if our stuff really really wants to phone the mothership to have Big Mulch pay us a visit, there are supposed to be ways for them to do that. Right?
___
EDIT: The topic we chose is also something that's not typically discussed in our social network, nor our kids' social networks. I will say that it's related to a profitable market, and we're in the target demographic, but we did our best to identify a market that we didn't have in common with our social groups.
It storms a lot here. And kids are notorious for not hearing things that are obvious to the rest of us. :)
My kids aren't paranoid, they don't freak out about the storms. It's just a natural thing, and a reminder that there's a potential that nature's best light show is in store for them. A nice side effect is that they're not afraid of storms—they look forward to watching them.
A similar one changes my LED-filled floor lamp (running WLED) to a pattern that matches the current weather conditions. Falling rain, lightning, heat "rising". Sort of an ambient notification system.
Another warns my wife if her commute time home is expected to be longer than usual, so she can opt to get a bit more work done if it means a quicker drive with a similar arrival time.
My favorite is the laundry notification. Current sensing outlets let us know when the washer or dryer is in use, and door sensors track if the lid to the washer or dryer is open/closed. So if someone starts laundry, the HomePods play a chime and announcement that "the washing machine is done!" If the door isn't opened in 15 minutes, it chimes again. Another 15 minutes, and a notification is sent to me and my wife. :D
Building are different in different countries, and other geopolitical divisions. The UK authorities, for example, say it's safe to shower during a storm as long as your plumbing has been properly integrated into a standards-compliant equipotential bonding, noting that said standards change over time. So, you'd have to know them and trust/verify the builder followed them.
Or, you know, just wait a while and enjoy the show.
The risk of death of showering during a lightning storm could be the same as the risk of death going for a 15 mile drive[0].
A typical lightning bolt is about 300 million Volts and about 30,000 Amps. Since even the best-grounded home certainly can't sink 30,000 amps of current into the neutral-bonded earth bar, it has to go somewhere. It finds multiple paths, and the current is shared between them. There's also the conversion to heat in all those insufficient conductors, etc.
{% set distance = states('sensor.lightning_detector_lightning_distance')|int(999) %}
{% set bearing = states('sensor.lightning_detector_lightning_azimuth')|int %}
{% set wind_bearing = states('sensor.pirateweather_wind_bearing')|int %}
{% set bearing_normalized = bearing % 360 + 360 %}
{% set bearing_left = (wind_bearing-80) % 360 + 360 %}
{% set bearing_right = (wind_bearing+80) % 360 + 360 %}
{% set distance_max = 20 %}
{% set bearing_min = ([bearing_left,bearing_right]|sort)[0] %}
{% set bearing_max = ([bearing_left,bearing_right]|sort)[-1] %}
{% set approaching = bearing_min < bearing < bearing_max %}
{% set close = distance < distance_max %}
{{ close and approaching }}
It normalizes the bearings to avoid dealing with the 0-degree crossover, and is reevaluated automatically whenever the three tracked entities change their state.
The automation itself is set to "restart" mode, and fires whenever the template sensor is true. It then waits for the sensor to be false for 15 minutes, and for the nearest lightning distance to be > 20 miles for 15 minutes.
Why a 20 mile radius? It's a good enough proxy for the time it takes them to prep for (get clothes ready, etc) and then actually shower (and brush teeth, 30 minutes total), and most storms in my area tend to move through at ~40mph.
As mentioned before, we don't worry. There's no judgment. It's a simple binary. And we spend the time enjoying the storm that we otherwise would've missed.
- “(The) honest answer:” (again, with colon)
- “The thing to internalize:”
- “The smoking gun:”
(really, sentences that start with “The <tag suggesting the next clause is the key point>:” are a strong tell, but those four are the most prolific)
- “load bearing” (when not talking about architecture)
- “blast radius” (when not talking about actual explosives, but rather the effect of an event/action)
- “smoke test” (esp. when “sanity check” is more apropos)
- Lists of three clauses/adjectives where the third is really just a combination of the first two
- Referring to the “shape” of things figuratively
- Social media posts that end with “Curious if anyone…”
- Stories or anecdotes using. “Oh. Oh.” (where the second “oh” is italicized)
Edit: Yes, some of those last ones are terms that we often use as devs...but I would argue about the actual frequency of their use. Plus, these tells live on in prose generated by the latest models.