Max Spero

@max_spero_ · 🌟 All-Star

2,678,200

Twitter Wars 2 points

Scores updated 13 min ago

Tweets: 153
Likes: 23,614
Retweets: 405
Replies: 602
Quote tweets: 94
2× boosted: 1
Followers: 14,244+844

Every scored tweet

300 per tweet · 100 per like · 200 per reply · 300 per retweet · 300 per quote tweet

Tweet	Likes	Retweets	Replies	Quotes	Boost	Points
this subreddit is a gem https://t.co/lcC2JbkXpJ Oct 18🥈 Silver	7,159	121	44	17	—	766,400
wait i think it's actually real https://t.co/gO1Jyh4OQ3 https://t.co/RalpHrPCmR Oct 22🥉 Bronze	4,108	30	48	5	—	431,200
it should tell you something that the most bleeding edge ai-pilled technologists are not letting AI write for them https://t.co/CJihCbwUwb Oct 13🥉 Bronze	1,611	102	56	14	—	207,400
oh, okay https://t.co/0x4pKd7pfK Oct 19🥉 Bronze	1,776	29	41	8	—	197,200
Today, we launched Pangram 4.1. We better calibrated Pangram's confidence, and we now show this tooltip indicating an AI range for documents we are less certain about. https://t.co/rMpCnyDprk Oct 19	914	43	52	13	—	118,900
@sqlanm We can, but we won’t Oct 19	709	18	27	12	—	85,600
i am going insane, what is going on here https://t.co/4xGdY9ffoX https://t.co/EocCW9e1BP Oct 21	600	1	42	0	—	69,000
I fear that any free API or hosted content will become untenably expensive to serve in the age of AI agents https://t.co/RGlPqvp3RT Oct 23	493	12	22	2	—	58,200
@entirelyuseles This was the prevailing advice in the comments Oct 18	562	0	1	0	—	56,700
Whenever I notice I am blocked by someone on twitter, I don’t take it personally or revenge block. Instead I wait for them to recognize their mistake and unblock me. I am now mutuals with 2 people who used to have me blocked Oct 12	412	3	36	5	—	51,100
how am i supposed to interpret my presence on such a list https://t.co/lYtvlLkhOm https://t.co/CrlTreeALS Oct 13	411	0	32	0	—	47,800
apparently canada is denying ISBNs to books primarily generated by AI https://t.co/bIMilTYdK3 Oct 15	329	18	15	9	—	44,300
@jannalulu_ the depth of shadow looks incredibly AI generated. but I think it may just be that one-in-a-million coincidence Oct 22	390	1	2	0	—	40,000
holy shit this is crazy technology https://t.co/OwrohutM4p https://t.co/37SgQrO4if Oct 21	323	6	12	0	—	36,800
seems this is AI? why does bezos* have two right hands https://t.co/QWJDjUh0wF https://t.co/fmcs5digIp Oct 15	201	1	23	2	—	25,900
wait this guy was actually serious https://t.co/V777wHwc0C https://t.co/k6WwErU1D8 Oct 13	188	0	4	0	—	19,900
@QualiaQuanta Hi! I'd like to take a look. Could you share the paper from 2016? Oct 21	191	0	1	0	—	19,600
@Rohanburdened oh okay she's just making a bunch of stuff up Oct 21	175	0	6	0	—	19,000
@DanAlroyPhilo @QualiaQuanta oh I see https://t.co/CMRTHTGk8m Oct 21	164	0	5	0	—	17,700
This is the only AI that Gen Z likes btw https://t.co/0G5Su6wG8h Oct 23	153	1	6	0	—	17,100
kind of messed up that clicking "unsubscribe" on the bottom of an email makes me click an additional 30 times to unsubscribe from all of their notification categories https://t.co/7j53AODiyo Oct 20	126	2	18	0	—	17,100
hmmmm https://t.co/zOkrjRW5b6 Oct 21	158	0	1	0	—	16,300
@QualiaQuanta I feel like I am going crazy, I searched Scope of New Mechanism and looked at all of the page one results. Is there something different I am not seeing? Can you just send a link? https://t.co/qYnYLScSPR Oct 21	139	0	1	1	—	14,700
@growing_daniel Not offended that some people still think pangram can't work. It goes against a lot of people's priors that AI text is detectable at all. Luckily, accuracy is one of the easiest claims to validate just by running a bunch of text with known labels. Oct 22	112	2	4	0	—	12,900
@QualiaQuanta is this the right one? I'm seeing 100% human written and also different text than the screenshot you posted. https://t.co/MHdFKVHdLT https://t.co/rkz6llpnq8 Oct 21	116	0	2	0	—	12,300
Pangram is hiring a Senior Brand Designer! We're looking for a talented designer to connect Pangram’s visual identity directly to our mission of becoming the voice of authenticity across the Internet. Please reach out if you're interested! https://t.co/8HxtfY6T6R Oct 22	85	7	5	1	—	12,200
We've seen confusion about Pangram's accuracy and how it related to length. Pangram is more accurate on longer text, and we wanted to demonstrate that through the UI. The results page is updated to indicate lower certainty for scans under 150 words. https://t.co/7hKYlgDWmn Oct 19	97	1	1	0	—	10,500
Pangram 4.1 also improves on some false positives that had been previously reported. In my experience, this model is a lot better calibrated and makes fewer real-world errors. https://t.co/HNWBzNVbmO https://t.co/X0jDC6Rskv Oct 19	76	1	3	1	—	9,100
Why are there SO MANY sophie bots now?? it's so eerie https://t.co/CBEMW1jJrV Oct 14	60	1	12	0	—	9,000
@mazeincoding That’s awesome. Fight slop with more slop Oct 15	80	0	1	0	—	8,500
@lulumeservey If you look at the quote tweets it’s ALL paid partnerships and they’re all slop accounts https://t.co/sUG9PNgirP Oct 12	71	0	3	0	—	8,000
@emweener https://t.co/eorGMinoa0 Oct 13	73	0	1	0	—	7,800
@AlexGodofsky this is also a wild exchange https://t.co/sZhpgW783G Oct 21	72	0	1	0	—	7,700
It took me approximately twenty times longer to order a pizza through a personal agent compared to doordash, but you really have to think of it as accepting a bribe to do the annoying onboarding tasks. The pizza is just a bonus https://t.co/Q78Xb8Sgps https://t.co/mF6ilTmNrh Oct 17	71	0	1	0	—	7,600
@mandylu boy do i have the tool for you Oct 12	70	0	1	0	—	7,500
@KhemianSilap yeah i will be the first to admit that. the text model has a couple years' head start in terms of development. but with that said I think it's pretty good! And getting better over time Oct 22	65	0	1	1	—	7,300
@shakoistsLog None of these words are in the bible Oct 12	68	0	1	0	—	7,300
@the_varunrana @pangram ai? Oct 20	60	0	2	0	—	6,700
Based. We already bank with Mercury but if I was starting a new company today I'd be trying to do as much as I could through them https://t.co/5Z0lpcPdW6 Oct 20	58	0	1	0	—	6,300
so the scammers gave up on impersonating bloomberg reporters and are on to impersonating openai employees? https://t.co/kWlhzHD2cb Oct 13	49	1	2	0	—	5,900
@saintcynosure Unclear that they are using any detector Oct 15	46	0	2	0	—	5,300
@zach_yadegari not a good look to interview someone and then publicly call their work slop. give direct feedback privately and be polite in public, not the other way around Oct 15	44	0	0	0	—	4,700
@sporadica On one hand, huge aura loss to buy a bunch of engagement and get near zero organic engagement On the other hand, at least the paid QTs are properly disclosed? Oct 13	39	0	1	1	—	4,700
@_chair i didn't make this account til i started pangram https://t.co/YYh8kDuAdb Oct 17	36	0	1	0	—	4,100
@justalexoki I think the whole image was generated. It’s got the openai lattice https://t.co/xKxvXm6X9R Oct 15	29	0	2	1	—	3,900
@sergeantsup @jannalulu_ Maybe I’m using the wrong term but like the contour of the light and shadow around the noodles smoothly spans the full range of bright to dark Oct 22	30	0	2	0	—	3,700
Model card with benchmarks here: https://t.co/QprVqeYBh9 Oct 19	32	0	1	0	—	3,700
it's just deep learning! you know how you can ask an LLM who you write like and, if you have enough public writing in the training set, it can guess who you are? it's like this but instead we're training a model explicitly on this task and it has seen millions of examples of claude, chatgpt, etc. writing in every domain imaginable Oct 22	28	1	1	0	—	3,600
@chaotictransfem i think genuinely it seems like they may be undifferentiated so the moat is that someone already invested time to sign up and enter their doordash credentials Oct 16	29	0	2	0	—	3,600
@benglickenhaus Just wait til they hear about Pangram 4 Oct 15	29	0	2	0	—	3,600
@powerbottomdad1 Guy who keeps getting divorced to goodhart this metric Oct 15	32	0	0	0	—	3,500
@eirircjsnsn yes :) Oct 19	27	0	2	0	—	3,400
I agree with staring into the abyss being an important skill, but sometimes people con themselves into inaction by refusing to do things that work today because they won’t work in the limit. You build insight today by working on adjacent problems so you can actually make a breakthrough in the future Oct 18	28	0	1	0	—	3,300
Interesting survey, fill it out if you have a couple mins https://t.co/OGpxWvrIrs Oct 14	21	2	0	0	—	3,000
AI that is more human-like is less verbose, less correct, less helpful, contains more expletives... if asked to write an essay it might say "no" or "i'll get back to you next week" all of this training pushes the model to diverge from the human distribution into something distinct Oct 22	21	0	2	0	—	2,800
Thank you @benglickenhaus for your hard work on this model! Oct 19	25	0	0	0	—	2,800
to give maximum benefit of the doubt, this may be true and the author is just very confused about it https://t.co/XvYGnTTW9S Oct 22	22	0	1	0	—	2,700
@ByrneHobart Lmao Oct 18	23	0	0	0	—	2,600
@elyasbuilds the tech side of streaming is commoditized, netflix is a media company now Oct 15	16	0	2	0	—	2,300
@benglickenhaus Gotta admit AI;DR is pretty good Oct 20	16	0	1	0	—	2,100
@ajhodls How about a stack ranking of who engaged with my tweets the most? Oct 20	14	0	2	0	—	2,100
@willdepue Thank you Will Oct 17	18	0	0	0	—	2,100
apparently it's ai upscaled. side-by-side https://t.co/U1x7Q2GwAa Oct 15	15	0	1	0	—	2,000
Oh my god https://t.co/L8wQUNbzPD Oct 23	14	0	1	0	—	1,900
@chaotictransfem the software looks pretty rough right now... most apps aren't updated for it Oct 23	9	0	3	0	—	1,800
@Duderichy It’s Luma Oct 20	15	0	0	0	—	1,800
@YoshkaMJ Instant reject for lack of character Oct 19	10	1	1	0	—	1,800
@CEBKCEBKCEBK No shot these two hands belong to the same person https://t.co/N54bbYLAhB Oct 15	11	0	2	0	—	1,800
@ted_dziuba @natashaghoskins @partiful https://t.co/jSPMJll9Ag Oct 12	13	0	1	0	—	1,800
The problem with any proposed fix to magic’s land system is that it homogenizes deck building. Magic has had standard decks that play 19 lands and others that play 28 lands. each enables a meaningfully different strategy and tradeoffs. If you simply got one land a turn for free, the range of viable constructed decks would be significantly more narrow. Number of lands is this measure of how much risk a player is willing to take in deck construction, but risk is only meaningful if taking too much has consequences Oct 16	12	0	1	0	—	1,700
@shitpost9000 maybe this one was for the better Oct 13	12	0	1	0	—	1,700
@Prigoose @chaotictransfem join the twitter wars Oct 12	12	0	1	0	—	1,700
@altryne @amitpr The path of least resistance to making text pass pangram is making it less and less coherent! Oct 18	11	0	1	0	—	1,600
@justalexoki Ask any vc to invite you to their screening Oct 17	5	0	0	0	2× Vibecamp	1,600 800 × 2
@mark_attar I like that Oct 13	11	0	1	0	—	1,600
@corsaren Yeah that’s fair. It’s annoying Oct 13	13	0	0	0	—	1,600
It’s probably a good thing that twitter made likes private https://t.co/3iJt2ajHK9 Oct 17	10	0	1	0	—	1,500
@smooth_normie I actually think there were a number of people who assumed pangram was another scam back when it was common belief that AI detectors were snake oil Oct 13	12	0	0	0	—	1,500
@_anmlfrm get the chrome extension! https://t.co/FBBHYH69Xg Oct 19	6	0	1	1	—	1,400
@lumirth Implying I don’t tweet during the work day Oct 16	9	0	1	0	—	1,400
@rogesterone @lfschiavo Crazy!! Must have been exciting to initiate conversation with the chinese agents Oct 14	11	0	0	0	—	1,400
@WilliamBryk insane trajectory Oct 13	11	0	0	0	—	1,400
@leaningaway Fewer than you might think. The pangram bot on the other hand… Oct 12	10	0	0	0	—	1,300
@birchlse thanks, it did seem weird that the PhilPapers abstract didn't match the abstract from the pdf Oct 22	9	0	0	0	—	1,200
@benglickenhaus 🎉🎉🎉 Oct 19	9	0	0	0	—	1,200
@ai_appreciator i don't like that Oct 13	9	0	0	0	—	1,200
@pleometric @norvid_studies @pleovision @DeanLearner Tell them it costs $1 per video and add a suggested donation to the iphone duo fund Oct 23	4	0	2	0	—	1,100
@amplifiedamp the pro plan does! https://t.co/TczOhexE6r Oct 20	4	0	2	0	—	1,100
@theojaffee Legend Oct 12	8	0	0	0	—	1,100
@Prigoose @desertgodhead @hermitfriend @funplings @__drewface @QiaochuYuan congratulations @QiaochuYuan, crazy run Oct 12	6	0	1	0	—	1,100
@granawkins Wow, that means a lot. Which ones did you already love? Oct 20	5	0	1	0	—	1,000
@notnullptr Keep on learning! Oct 19	7	0	0	0	—	1,000
@emweener I’ll let you know on Saturday Oct 15	5	0	1	0	—	1,000
@chaotictransfem https://t.co/3Gqpso9vN1 Oct 14	7	0	0	0	—	1,000
@HououinTyouma Meta AI augmented reality glasses will come for these jobs too Oct 13	7	0	0	0	—	1,000
@Bonecondor One day, surely Oct 13	7	0	0	0	—	1,000
@Nition @growing_daniel there was ~120 words of a tale of two cities that pangram 4 classified as AI https://t.co/yUTQ403fll Oct 22	4	0	1	0	—	900
@notgodcomplex96 @EntropyChase @JaredWalczak are you using a copycat website? otherwise i'll pull out the $1,000 bounty if you have a book written 20-40 years ago classified as 90% or more AI Oct 21	4	0	1	0	—	900
@vikhyatk @zacharynado LFG congrats! Oct 13	6	0	0	0	—	900
@emweener @pangram What can I say, I have enough going for me that I don’t need to constantly talk about being w*sian Oct 13	4	0	1	0	—	900
@jtevesobs Oh this is good Oct 13	6	0	0	0	—	900
@_ueaj First personal agent that can control my phone will rule them all Oct 17	3	0	1	0	—	800
@benglickenhaus this is a lot of things honestly https://t.co/H43mHtNAtA Oct 17	5	0	0	0	—	800
@SHL0MS oh thanks for finding this, that's crazy, the upscaler changed so much. bag texture and shorts are completely different too. eerie Oct 15	3	0	1	0	—	800
@transkatgirl I pay >$100 for @screenstudio even though there are probably free OSS tools that are functionally equivalent Oct 15	5	0	0	0	—	800
@rawandferal yoooo how was the class? Oct 13	3	0	1	0	—	800
@emweener https://t.co/I2yLv0wRg3 Oct 13	3	0	1	0	—	800
Thank you PLEOVISION https://t.co/4Jq2fEAjby Oct 23	4	0	0	0	—	700
@bryancsk it's so true Oct 23	4	0	0	0	—	700
@crystalxduan The public wanted to get fooled? Oct 22	2	0	1	0	—	700
@calxoxo Thanks!! Oct 19	2	0	1	0	—	700
@aliceisplaying ty ty! Oct 19	4	0	0	0	—	700
@ByrneHobart My apple intelligence told me to say “banger” Oct 16	4	0	0	0	—	700
@TuhinChakr @RossBarkan I’m seeing it tomorrow! Got tickets early, theater is fully sold out it’s crazy Oct 16	2	0	1	0	—	700
@lumirth @navyadotbio It’s kind of just where i was when i started the company Oct 16	2	0	1	0	—	700
@pleometric @norvid_studies mmmm banana bread Oct 13	4	0	0	0	—	700
@pleovision explain this post Oct 23	1	0	1	0	—	600
@__venki__ @pangram We do ZDR for enterprise but yeah i get that it’s hard if it’s not top-down Oct 23	3	0	0	0	—	600
@MaleManlpulator 🙏 Oct 22	3	0	0	0	—	600
@Nateemerson send tweet Oct 19	3	0	0	0	—	600
@aliceisplaying And that's on me. Oct 19	3	0	0	0	—	600
@GabGarrett @whatiwrote_ @ednewtonrex they should ban any writers who lift most of their book from google search results Oct 17	3	0	0	0	—	600
@sevensix43 @mazeincoding I assume they are using fraction_ai Oct 15	3	0	0	0	—	600
@trashpandaemoji @GergelyOrosz it exists! https://t.co/6bHLkGT4rL Oct 15	3	0	0	0	—	600
@notpurav Not really https://t.co/RRFwrsphID Oct 13	3	0	0	0	—	600
@theRealJohnPeng sure Oct 13	1	0	1	0	—	600
@amankabeer Nothing to gain Everything to lose Oct 12	1	0	1	0	—	600
@pleovision explain this post Oct 23	0	0	1	0	—	500
@simonsarris that's hilarious and also tragic Oct 23	2	0	0	0	—	500
@AfterDaylight oh congrats! I'm not too familiar with pedagogical theory but seems like there is a path for AI use without harmful cognitive offloading. but it starts with inspiring a desire to learn in the students which seems WAY easier said than done. Oct 20	0	0	1	0	—	500
@IsaacKing314 Thanks for all the feedback! Oct 20	2	0	0	0	—	500
@notnullptr @gergomoricz @MikeyWhite189 That’s hilarious lmao Oct 18	2	0	0	0	—	500
@Bonecondor the contacts were even cooler in person Oct 16	2	0	0	0	—	500
@lumpenspace @RuxandraTeslo @growing_daniel Looking forward to it Oct 22	1	0	0	0	—	400
@andersonbcdefg @elyasbuilds come visit Oct 21	1	0	0	0	—	400
I'm always happy when people test Pangram and write about it https://t.co/BOXZUuXjs6 https://t.co/rqiepb5RCx Oct 20	1	0	0	0	—	400
@granawkins nice! enjoy. I thought the whole 9 video games thing was fun, I'd like to see more of other people's music picks Oct 20	1	0	0	0	—	400
@RoryWalshWatts thanks man! Oct 19	1	0	0	0	—	400
@deepfates @_chair 2021!! Oct 18	1	0	0	0	—	400
@gpt972394 Is that a Gen Z thing? I’m seeing grown men on twitter sniped by rule 63 Oct 23	0	0	0	0	—	300
@PanPanworkshere Oct 23	0	0	0	0	—	300
@EigenGender @growing_daniel or out of distribution text, I'm easy to please Oct 23	0	0	0	0	—	300
@AfterDaylight and assessment still needs to be rigorous, I keep seeing bad outcomes as a result of increasingly lax assessment. Oct 20	0	0	0	0	—	300
@IsaacKing314 Pangram already does pretty well on code but official support would mean a much bigger investment into evals. Not sure there is large demand for it Oct 20	0	0	0	0	—	300
@ajhodls It’s only fair Oct 20	0	0	0	0	—	300
@majamediaco thanks maja! Oct 19	0	0	0	0	—	300
@ADLXBT Just one Oct 19	0	0	0	0	—	300
@MilesBuilds long docs can land a range if our uncertainty range is larger than 4%! Oct 19	0	0	0	0	—	300
@e0d40e74 Ah you’re right Oct 15	0	0	0	0	—	300
@transkatgirl @screenstudio there* oh no, cognitive decline Oct 15	0	0	0	0	—	300
@bernhardsson plant on a chair 🔥 Oct 14	0	0	0	0	—	300
@rayzhudev No idea it just happened eventually Oct 13	0	0	0	0	—	300
@rayzhudev 6 months or a year? It’s hard to say exactly when but suddenly they showed up on my timeline and I could reply Oct 12	0	0	0	0	—	300
Total · 153 tweets in Twitter Wars 2	23,614	405	602	94	1 boosted	2,678,200

Posts, replies, and quote tweets count — native retweets are excluded. Mentioning a project on the boost list doubles that tweet's points.