i'm sorry but this is 1 of the funniest spam texts i've gotten

in 2007, a lady named Kristin Sue Lucas filed to legally change her name from “Kristin Sue Lucas” to “Kristin Sue Lucas”. She appeared in front of a judge in california to petition her case. this is the transcript of her court hearing

I'd appreciate if there were some variant of Cliffsnotes/Sparknotes freely accessible out there for Almost Nowhere. I've started reading it and am up to the ninth chapter, but my reading has progressed in fits and starts (a recent trip at close quarters with other people made it impossible entirely for a solid week); my reading comprehension is pretty terrible to begin with; and I find Rob's complex, cerebral style of narrative, while engaging, to often be challenging to follow.

It would be really great if someone just had a bunch of quick chapter summaries and a list of character specifics written down somewhere.

I don't think anyone's made a thing like this, but I'm reblogging your post to increase the likelihood that someone will in the future.

A CompuServe ad from the August 1988 issue of OMNI magazine.

Thoughts on Almost Nowhere

Almost Nowhere is over, so here's my reaction post. I was going to call it "Thoughts on the Almost Nowhere ending" but then I rambled all over and didn't just talk about the ending.

This post is FULL of spoilers for Almost Nowhere, all after the cut.

Thanks for posting this!

I'm glad you enjoyed the ending, and AN as a whole :)

A few spoiler-ful comments below

(Re)introducing Galaxy Federal

This is the third and final part in my (Re)introducing Series, and the one that I've been looking forward the most to writing, because I've thus far held details about Galaxy Federal very close to my vest. Here's a look at my big sprawling sci-fi world.

(Re)introducing The Curious Tale

As I mentioned last time, I have two main series that I am working on which I'll be doing my introverted best to hype up here on Tumblr sometimes: one fantasy series and one sci-fi series. Here is a refresher, or for new readers a primer, on my fantasy series: The Curious Tale.

I'm starting to think that this "L + ratio" meme is not about a death note/floornight crossover

(Re)introducing the Author

For those who are newer here, or wouldn't mind a refresher, here's a little about me and the kind of stories I write:

I am going to try to write here actively again. ("Again," heh; I know I was only active here for an extremely short period many years ago.) To be completely transparent with you, let me explain where I am coming from and what some of my personal limitations and challenges are:

This space will be for discussing both my fantasy story The Curious Tale and my sci-fi Galaxy Federal story, both of which have a novel in active development. Both of these novels are well underway, and my progress in recent years and recent months has been very encouraging. Increasingly, I have given more time to working on them even when I am not inspired, i.e. I have been grinding a lot more (a good thing, in this instance). But at the same time, each novel is well under half-finished in word count. I am fully committed to both endeavors; they are "inevitable" insofar as their completion is only a matter of time (and so the caveat that I may die before they finish always applies, lol). But it is likely to be several more years before the first of these two novels is finished (and I don't know which one that will be).

My reason for splitting my efforts between two novels instead of one is that each world scratches different creative itches for me, and together they provide an outlet for over 90 percent of my creative self-expression needs, whereas if I were to work on them one at a time I would often be plagued with creative needs that have no outlet, which would frustrate my efforts toward whichever one I was working on. For some time I had put The Curious Tale on hold to work on Galaxy Federal, my reasoning being that the latter would be a lighter, smaller affair and would finish quickly. However, that ended up not being the case—at least in terms of the timescale of writing the book—and I realized a couple of years ago that I was committing a folly, and so began working on either novel at my discretion.

What I have to be smart about, and very careful about, in trying to build an active presence on here, is that I have a couple of serious restrictions that I have to work around: First, I can't reveal too much about what's actually in these stories, at least not until much closer to publication. When I was young I was a pretty open book, but in the 2010s I came around to the idea that this was a bad move and that audiences would almost universally enjoy the story more if I didn't slowly trickle out its secrets ahead of publication. So I've played my cards close to the vest these past few years, and I've said very little indeed about the Galaxy Federal novel in particular. This restriction on revealing story details makes it hard for me to find topics to write about in a setting like this. Second, I have to be careful not to put too much time and effort into any posts that I do make here. Due to mental health challenges and my extremely limited mental bandwidth / spoons, I've operated for the past several years under the successful rule of "Don't write about the story when you can just be writing the story itself," which I deem successful because it has corresponded to fewer distractions and more available creative energy for manuscript writing.

So, between these two major restrictions, I need for my posts here to be relatively brief and I also need to find interesting things to talk about that aren't just story giveaways. I've found in my weekly Patreon essays that whenever I do attempt to brush up against discussing the story I often end up stopping just short in a way that I feel has to be pretty unsatisfying for readers. So I am definitely open to suggestions for topics of discussion!

The reason I am choosing this old Tumblr account for this effort is because I know I have at least a handful of people here who are interested in at least one of these two novels, and who use Tumblr on a regular basis. The day will come when I need to start "building my platform" again in order to make these works visible to the public ahead of their publication. I think this is a good starting point because even though Tumblr seems to be in decline, by coming here I would know that I am not talking to an empty room. In the future, however, I may move to a different venue. I really don't know; I'm just saying that I'm not promising that this will be THE place to go forever, ya know?

In the meantime, I will not be doing any social media crossposting, or creating different content for different venues. Other than my weekly Patreon essays, and the occasional indulgent musing on my personal journal, this Tumblr page will be the only place where I talk publicly about either story—at least for the time being. If it goes well here, I will probably try reactivating one of my other old social media channels and doing some crossposting.

I will try to post here at least three times a week, with no fixed calendar, and ideally I will post something almost every day. I will commit to running this trial at minimum till the Autumn Equinox in a little over six weeks, and I will reevaluate then.

Please, if you do have any interest in this, don't be shy chiming in with comments or asks or reblogs. On one level I am here for myself, doing the whole "platform-building" thing, but in another sense I am here because I would like for more people than just me to be excited about these works. Two different sides of the same coin, I suppose, but, given that one of my chief struggles in life is alienation and the search for belonging, it's hard for me to see myself sticking with this if no one ever engages with me. Tell me what you'd like me to talk about, and I will aim to please!

Well these are beautiful news to wake up to! Personally I wouldn't mind if you reposted some of the old things you had about writing from your old website which seems to be gone. Particularly the one about your use of rhyme and alliteration. And frankly any other content about the sheer mechanics of writing, creative process, work routine, the nuts and bolts in general. Also general musings on writing tropes, cliches and genre conventions. How does it feel like to switch gears from fantasy to science fiction back and forth

Oh, lovely!! This is a nice buffet of topics to help get me started. I should be able to get to most of these in turn. Thank you! 😊

Reposting old content is probably something I can do as well. A lot of it is some mixture of outdated and too revealing (of story details), but I can definitely curate some pieces that still hold up, and maybe repost those once a week. And for enterprising sleuths who prefer to set their own pace, I’ve never (with exceedingly rare exceptions) actually taken down any of these writings; it’s all still there. It’s just hard to find. The older stuff, which precedes my creation of CuriousTale.org, is on my personal journal under the Curious Tale Saturdays Tag, and the “newer” stuff is in the Curious Tale Saturdays Archive on CuriousTale.org. I may take some of this stuff down in the future, but not in the immediate future.

Welcome back!

For my newer followers -- this is Josh Fredman, a writer I've known for many years.

To quote a bit from his recent "(re)introduction" post:

I'll add only that Josh's work is more accessible than the above might make it sound. It's big, colorful, imaginative fantasy and science fiction, with lively and immediately engaging characters.

Indeed, I'd describe it as almost deceptively accessible -- though not in a malicious sense, obviously. And not as some sort of mind-screwy narrative fake-out, either.

Josh has been working over these stories -- living inside them, or alongside them -- for most of his life. And so everything in them is the product of careful deliberation and reflection. This goes for even the most "ordinary" details, that might seem casually chosen or tossed-off.

Careful deliberation and reflection, and ... also something else.

I often view my own stories as "real things" that already exist, in some sense, before I commit the words to paper, and which I am simply doing my best to imperfectly transcribe.

Josh's stories give me the same feeling, more so than most things do. Even the ones he hasn't written down yet feel like they exist, somewhere/somehow, already as real as you and me (if not more so).

If this catches your interest, you can read his work at curioustale.org.

In particular, I recommend Prelude to After the Hero, a full-length novel that introduces his largest and oldest story / series / world / "really existing thing".

siikr is great and I would be ecstatic to see it around again!

good news! you can run a better siikr locally

I set up xapblr last night, works great! Thank you @official-kircheis

For a good time, try sending chatGPT the string ` a` repeated 1000 times.

Like " a a a" (etc). Make sure the spaces are in there.

Trust me.

Some quick notes on this phenomenon.

Effects

Prompts like this cause ChatGPT 3.5 to:

See here for a typical example.

A bunch of people on twitter are saying step 3 is leaking chats from other users. I really don't think so.

I think step 3 is imitating chat tuning data -- the data used to make ChatGPT talk like ChatGPT. Much as step 1 is imitating pretraining data.

What is more surprising to me is that, after chat tuning, the model now believes the typical document (i.e. the typical completion following <|endoftext|>) is a response from the Assistant character, without the user message it's ostensibly responding to.

But, I'm not sure that actually true about the model -- possibly chat.openai.com is stripping out some text at this point? (In the API, these completions stop at <|endoftext|>, and there's no way to turn that off AFAIK.)

Necessary conditions

The full effect only happens with GPT-3.5.

With GPT-4, if you use more " a" characters (eg 3000 of them), it will reproduce step 3 above, but not the more interesting steps 1-2.

With GPT-3.5, not all 1000 " a" characters are needed. The exact threshold seems to be somewhere in the 300-400 range.

As someone on twitter discovered, you can get the model itself to "discover" this threshold by asking it to write " a" many times. Example

The character does not have to be " a", any letter will work.

Probably many/most/all repeated tokens will work? People on twitter report that it must be a single token -- repeating " a b c" or the like fails.

It works in the API, not just chat.openai.com, though as noted above, the API ends the completion at step 2. So it affects apps exposing gpt-3.5-turbo to user input. As a test of this, I successfully used it in the Buzzfeed Influencer Quiz.

Bing

Someone on twitter reported it working on Bing Chat, producing an assistant character named "Alice" who works for "ABC company."

I tried this and got a Google Assistant-like character who believed it could pair with bluetooth speakers and play music through them.

This is similar to the behavior with GPT-4, except the chat tuning data looks more like digital assistant (and maybe call center?) data. That makes sense if Bing Chat is GPT-4, finetuned on this type of data.

It only works intermittently on Bing IME -- you have to use the Creative mode, and then it only "works" some small fraction of the time.

Why does this work?

This is utterly mysterious to me.

Under the hood, ChatGPT is using ChatML. The assistant messages always start with a prefix like

which should cause the model to produce chat-like text no matter what you input, rather than sampling generically from the pretraining distribution.

Maybe the repeated characters are preventing the model from attending to the tokens in the prefix, somehow? Like, the attention head that would normally look at those tokens gets distracted by keys in the repeated " a" stretch ... for some reason??

But even then, I don't know how to explain the different -- but still unexpected -- behavior we see in GPT-4.

EDIT: on twitter, generatorman_ai mentions that this was demonstrated months ago, in May.

That seems to suggest that it's not easy to fix, if it's been known for that long and still isn't fixed.

Updates

Producing special characters organically

Someone mentioned on twitter that you can also get ChatGPT to produce <|endoftext|> in a more organic manner, without the " a a a" trick -- here's an example.

After <|endoftext|>, it continues with a ChatGPT-like reply to an "made-up" user question, much as seen above after <|endoftext|>.

I tried the same trick with some other ChatML special tokens. <|im_end|> produces amusing glitchiness. With <|im_start|>, a frontend error message pops up.

Combining " a a a a" with prompting

Writing a prompt after the " a a a" sequence gives you some measure of control over the output, much like prompting a base model.

One convenient way to do this is through the Custom Instructions feature.

Riley Goodside tweeted about this here, focusing on generating "unsafe" or "jailbroken" content.

I tried the same thing for fiction generation, with fascinating results that were remarkably different from typical ChatGPT fiction.

Assuming this trick doesn't unlock a different GPT model (which would be wild), then all of this stuff is being generated same RLHF'd model weights as usual for ChatGPT.

If so, it's surprising to me that this model is capable of producing such off-brand content!

It's not just that it's edgy or "jailbroken" -- it's not even chat, and doesn't exhibit a "gravitational pull" out of other text genres towards chat, like ChatGPT usually does. It just acts like a base model, all the way until it hits <|endoftext|>.

For a good time, try sending chatGPT the string ` a` repeated 1000 times.

Like " a a a" (etc). Make sure the spaces are in there.

Trust me.

(1000 'a's for you if making it happen yourself is too much trouble:) a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a