[HN Gopher] The funny rules of SpamAssassin in 2023
___________________________________________________________________
The funny rules of SpamAssassin in 2023
Author : alexis2b
Score : 177 points
Date : 2024-01-02 09:27 UTC (13 hours ago)
(HTM) web link (updown.io)
(TXT) w3m dump (updown.io)
| lloydatkinson wrote:
| Some of those highlighted rules, such as using CC or having the
| string "can help" being used to decide if something is spam or
| not is so absurd I'll make sure to never use SpamAssasin.
| jeffbee wrote:
| It is not and has never been a good classifier. If open AI fans
| want to contribute something of value to society, they would
| train a spam classifier on a large, manually-labeled corpus of
| mail, where the features include envelope data. That would get
| open source maybe 10% of the way to Gmail quality, or 100x
| better than SA.
| CamperBob2 wrote:
| Does SpamBayes still work?
| dugite-code wrote:
| The Python project that's clent side? Best of my knowledge
| it hasn't been updated in years, spamassasin's in built
| Bayesian classifier works just fine.
| jepler wrote:
| I still used spambayes up until I basically abandoned my
| self-hosted e-mail setup (in favor of an @gmail).
|
| It occurred to me recently that LLM-style tokenization +
| bayesian classification would be a sweet upgrade for
| spambayes, which always struggled with ad-hoc
| tokenization rules.
|
| (I don't think of it as "client side"; it was integrated
| with my system via procmail on what I'd call the "server
| side". You could use it in other ways, including as an
| Outlook plugin, way back in the day. Or it could connect
| to an IMAP mailbox and filter messages it found, etc.
| Really versatile tool for its time)
| acidburnNSA wrote:
| FWIW I've been using SpamAssassin for over a decade
| personally (partly to avoid Google dependence), and it's been
| pretty darn good once I ran the Bayesian learning thing a few
| times many years ago. I get like 3-5 spams per week in my
| inbox. Do others really consider SA that bad?
| brokensegue wrote:
| On Gmail I get maybe 1 spam a month max in my inbox (and it
| blocks many per day)
| acidburnNSA wrote:
| That's pretty good. I have no doubt that Gmail spam
| protection is better than my self-hosted SA protection.
| For me, independence from a somewhat suspicious large
| company for something as important as e-mail is worth it.
| dugite-code wrote:
| I have a gmail honeypot where I fetchmail junk email
| straight to my junk folder and have a scheduled sa-learn
| cronjob. Ever since I started this I essentially stopped
| getting junk email in my selfhosted inbox.
|
| I also have dovecot set to learn Ham every time I file an
| email from the inbox to a folder for good measure.
| kjs3 wrote:
| So...statistically insignificant difference from SA for
| most mail users.
| brokensegue wrote:
| 3-5 spams per week vs. 1 per month is a big difference
| kjs3 wrote:
| I got (if I'm reading this right) 5500 emails
| (junk+delivered) to my personal mail account from
| 12/01/2023 to 12/31/2023. So that's a minimum, since I
| don't see the ones that get flushed out before I even see
| them.
|
| 1 spam a month would be .018% of emails and 5 x 4 spams a
| month would be .364% of emails
|
| So I would have gotten about .346% more spam based on the
| number of emails. In reality, because I don't see all of
| the mails, it's less. Is a touch more than a third of a
| percent a 'big difference'? YMMV.
| notatoad wrote:
| > I get like 3-5 spams per week in my inbox
|
| I'm more curious about the opposite metric: how many non-
| spam emails a week arent getting delivered to you? Because
| that seems to be the real flaw in spamassassin: the false
| positive rate.
|
| And the spamassassin users don't usually have much
| visibility into this, so when emails don't get to them they
| just blame the sender.
| JohnFen wrote:
| > how many non-spam emails a week arent getting delivered
| to you?
|
| My false-positive rate is very low, maybe a couple per
| month. However, I can predict with a high degree of
| accuracy when a piece of email is likely to land in the
| spam folder. Things like confirmation emails,
| registration emails, etc. are guaranteed to land in the
| spam folder. It's pretty hard for any system to
| accommodate those without allowing spam to get by.
|
| That's fine by me, though, because I know when to check
| my spam folder.
| arp242 wrote:
| FastMail uses SpamAssassin, and I get less than one spam
| emails in my inbox a month, with essentially zero false
| positives (which is the tricky bit where gmail seems to
| fail - I'd rather have the occasional spam in my email than
| false positives).
|
| In short: you can probably do better than 3-5 spams per
| week with SA.
|
| The big problem is the entire thing is a beast to configure
| with all the documentation of a Babylonian cuneiform stone
| tablet.
| CableNinja wrote:
| I hate to agree here, but configuring SpamAssassin is
| pretty rough. That being said, once its done, its pretty
| bulletproof and doesnt usually require messing with it
| all the time.
| JohnFen wrote:
| I find SA works excellently, personally.
| jerf wrote:
| We'll get there eventually, but it will be a bit. Spam
| classification at scale is already a compute-bound, or at
| least compute-starved, operation. Spam classification systems
| already do what they can to avoid so much as invoking a virus
| scanner if they can avoid it, because at scale it's so
| expensive. LLM-based spam classification is another order of
| magnitude more expensive _and_ would require hardware that
| current spam systems do not have.
|
| But that's a problem that will resolve itself over time, in a
| variety of ways. And the spam systems can play the same
| tricks with only invoking it on a fraction of emails too, of
| course. It's just at current expense levels, that would be a
| very small fraction indeed. I'd hazard that trying to use
| modern AI on spam classification at scale could easily
| consume 10x-100x of _all current AI hardware_ and still make
| less of a dent than you 'd hope.
| jeffbee wrote:
| It doesn't need to be computationally costly because, as
| you seem to imply, there are tiers of cost tradeoffs. You
| can invoke a very cheap classifier at SMTP time, that is
| biased to have few false positives, that will temporarily
| reject all that which is highly likely to be spam. You can
| do this without even glancing at the body. Of course,
| having signals about peer reputation is the strong suit of
| Gmail or Microsoft, and the distributed, open community
| would need to solve the problem of promptly updating and
| distributing such reputation signals. And by "promptly" I
| mean within seconds of the leading edge of an attack.
|
| Then there are increasing tiers of cost that you would only
| run after it becomes likely that the message is acceptable.
| As you say, you would only run an antivirus on a message on
| the verge of delivery, because decoding the attachment and
| running the AV (in an expensive sandbox) is so costly.
| dugite-code wrote:
| Really any of the open source language models might work well
| enough for the job. If you could manage to get a classifier
| that runs with tensorflow to take advantage of a coral tpu it
| would certainly be a major step up with managable
| performance.
| thesuitonym wrote:
| I hope against hope that AI spam detection never becomes a
| thing. At least with today's methods, I can tell a person why
| their message was marked as spam. If AI detection becomes the
| norm, all I can do is shrug and say, "Sorry, it's the
| algorithm."
| potatoman22 wrote:
| How do you think SpamAssassin/gmail/outlook created their
| spam rules?
| jeffbee wrote:
| Gmail has used machine learning to classify spam since its
| creation.
|
| https://workspace.google.com/blog/identity-and-
| security/an-o...
| Mailtemi wrote:
| Gmail is successful because it naturally is the biggest
| honeypot. Most antispam API filters are like
| accumulators. When a trend is detected, the rest are
| protected. But overall, it's about scale.
| layer8 wrote:
| SpamAssassin has a Bayes filter that you can train with ham
| and spam. This has basically been a thing since forever.
| andrewf wrote:
| "A plan for Spam" (2002) -
| https://paulgraham.com/spam.html
| El_RIDO wrote:
| Correct, Spamassassin will not classify emails be default.
| But it does include a Bayesian classifier and tooling that
| can be used to train it with curated ham and spam emails. It
| does require extra steps to set this up and feed it selected
| maildirs (for example, exclude inbox, spam and trash for
| training ham).
|
| https://spamassassin.apache.org/full/3.0.x/dist/doc/sa-
| learn... https://cwiki.apache.org/confluence/display/spamassa
| ssin/Bay...
| creeble wrote:
| I find it far more likely that AI will be used (indeed, is
| used already) to generate spam, rather than filter it.
| throw0101d wrote:
| > _being used to decide if something is spam or not_
|
| Each rule has a score associated with it. By default a message
| needs to reach 5.0 to be marked as "spam":
|
| *
| https://spamassassin.apache.org/full/3.0.x/dist/doc/Mail_Spa...
|
| The threshold is configurable. An header is added post-
| processing, e.g.: X-Spam-Status: Yes,
| score=21.6 required=4.0 [...]
|
| *
| https://cwiki.apache.org/confluence/display/SPAMASSASSIN/X+S...
|
| One can then choose what do to with this information (via
| procmail or Sieve). There is another header as well:
|
| > _X-Spam-Level: This displays your spam level with asterisks,
| with one asterisk displayed per point, rounded down. For
| example, if your overall SpamAssassin score is 4.3, it will
| display ****. If you score less than 1, for example, 0.5, it
| will display nothing._
|
| * https://www.mailercheck.com/articles/spamassassin-score
| londons_explore wrote:
| Having the rules public seems to take away most of the
| benefits...
|
| Any smart spammer will just tweak his spam to not hit these
| rules... And if he hasn't, it's because the vast majority of
| people don't use SpamAssassin
| Hnrobert42 wrote:
| >smart spammer
|
| I am sure there are plenty of smart spammers, but it also seems
| like a lot of spam comes from folks using scripts and email
| lists they use without fully understanding. It appears
| SpamAssassin would help with those operations.
| bombcar wrote:
| I'm starting to think the smart spammers are the ones selling
| worthless spam tools to the dumb spammers, because so much
| spam is entirely unactionable.
| brightball wrote:
| Part of the smart spammer approach is to condition people to
| what spam looks like, so you're more likely to let through
| the ones they really care about.
| adrienjarthon wrote:
| Hi, Adrien here author of this article (and of updown.io). That
| is true and I actually hesited to write the article for this
| reason, because it could make the spammer life easier. But
| after seing some of the legacy and nonsense in here I though
| it's still worth it so people at least understand what they are
| using.
| bell-cot wrote:
| > Any smart spammer will just...
|
| Spam is all about high-volume/~no-cost delivery of crap. Time
| spent tweaking the spam - to evade $Defense_1, $Defense_2, etc.
| - is added cost. Especially if $Defense_n is only used by a few
| of the prospective victims (folks too savvy or paranoid to be
| suckered _do not count_ ), then tweaking to get around
| $Defense_n is a losing strategy for the spammer.
| kjs3 wrote:
| _it 's because the vast majority of people don't use
| SpamAssassin_
|
| Bingo. Not that there aren't a lot of people running SA, but
| spammers want to be able to deliver to the big players(1)
| (gmail, o365, etc), not the size folks out there running SA.
| It's not worth their time to devote effort to optimizing for a
| rounding error in the deliverability equation.
|
| (1) Unless they're selling 'targeting' services where you're
| paying to deliver to a specific domain/user which might be
| behind SA. Plenty do, but that's a little bit farther down the
| criminality spectrum and vastly less volume than shilling
| peener pills or warranty extension scams.
|
| edit: formatting
| layer8 wrote:
| It's effectively no different from a spammer running their spam
| mails through a local SpamAssassin to see how it goes, and
| tweaking them until they pass.
| chrismorgan wrote:
| The _vast_ majority of operations in these fields are mind-
| bogglingly simplistic.
|
| Well-known rules will block most spam, some with occasional
| collateral damage but many with no realistic chance of
| collateral damage.
|
| Entity-encoding @ as @ in email addresses in HTML will
| block the vast majority of email address harvesters, with no
| collateral damage.
|
| Adding a honeypot field to an HTML form, with the label "If you
| are human, leave this field blank" and hidden by CSS, will
| catch practically all spam submissions, with no collateral
| damage.
| Anthony-G wrote:
| Spammers seem to be really lazy. The only time I've received
| spam on my own domain was when I changed servers and forgot to
| enable the Postgrey service. Grey-listing has been around for
| long enough (a couple of decades) so I would have expected
| spammers to be resending emails that are rejected with a
| temporary error.
|
| So I wasn't expecting Postgrey to provide much benefit. As it
| happens, in 10 years of running my own mail server, it's the
| only anti-spam measure I've had to bother with.
| njharman wrote:
| Spam is a minimal effort endeavor. A lead generator for scams.
| Only 0.1% or 0.01% response rate is good. They only want the
| unsophisticated, naive recipients. Expending effort to tweak
| rules is not worth the effort. (Although if they can get AI to
| do it, then that's back to near zero effort)
| srmarm wrote:
| I've been using SpamAssassin for at least 15 years and it's sadly
| gotten less useful as the spam arms race has moved on. We
| regularly see people on here post about deliverability issues
| with Gmail/Outlook but the truth is that sender reputation is by
| far the biggest indicator of whether a message will be spam -
| these type of rules are just counting deckchairs on the titanic
| in comparison.
|
| And this plays into the strengths of the big mail networks in
| detection. It's a bonus to them that every time they block a
| smaller host there is a good chance that sender will consider a
| move to office365 or Google Workspace for their mail.
|
| As an aside, not sure if OP is related to them but updown.io is a
| nice service and I appreciate the simple PAYG pricing! For what
| it's worth their mails seem to get through successfully to me
| too.
|
| Also for those facing mail delivery issues (or just practicing
| good email hygiene) - I recommend www.mail-tester.com - they give
| you an email address to send a mail to and carry out a heap of
| tests - including checking against SpamAssassin + blacklists,
| SPF/DNS/etc testing.
| andrewfromx wrote:
| there needs to be like a mozillia vs chrome thing here no?
| What's the best try so far for something like letsencrypt or
| mozilla foundation for not owned by big tech email so "will
| consider a move to office365 or Google Workspace for their
| mail" the sender has this other awesome option?
| jeffbee wrote:
| If you wanted to operate a haven for independent email
| hosting, where you want to assure deliverability in the face
| of Gmail's sender reputation system, you would need to
| classify your outbound traffic, and have a death penalty for
| spammers. If you tolerate any activity that peers classify as
| spam, that would tank your reputation.
| BSDobelix wrote:
| I like rspamd much more (performance and redis) than
| SpamAssassin, and as you mentioned:
|
| -https://www.mail-tester.com
|
| -https://www.learndmarc.com
|
| -https://mecsa.jrc.ec.europa.eu/en/
|
| Are exellent tool's to check your "deliverability".
| CableNinja wrote:
| Suprised to not see https://mxtoolbox.com in this list too
| Retric wrote:
| Reputation works well because of those other rules. If every
| office365/gmail email got through and everything lose was
| blocked spammers would just move to those platforms. Thus email
| inspection is a critical component enabling reputation based
| filtering.
| rlpb wrote:
| > It's a bonus to them that every time they block a smaller
| host there is a good chance that sender will consider a move to
| office365 or Google Workspace for their mail.
|
| The irony is that a substantial amount of the spam I receive
| comes from those platforms.
| nradov wrote:
| Are you certain the spam is actually coming from IP addresses
| controlled by those platforms? It's common for spammers to
| fake the SMTP headers.
| bombcar wrote:
| Quite a bit is "genuine", at least with Gmail, because
| infinite monkeys can sign up for infinite accounts.
| ceejayoz wrote:
| I get plenty of spam from Gmail accounts with SPF and DKIM
| passing.
| Mailtemi wrote:
| I've got sometimes a legitimate Google or MS dev
| newsletter emails going into their own spam folders :) .
| 8organicbits wrote:
| I've seen mail from my work Google Workspace that I sent
| to my own personal Gmail get flagged as spam. It's me
| sending to me. Google to Google. Logged into both account
| on the same computer.
|
| If anything I'm nervous to recommend Google because they
| flag too many legitimate emails as spam. After years of
| not checking, I'm checking spam again.
| cj wrote:
| > mail from my work Google Workspace that I sent to my
| own personal Gmail
|
| Does your company do outbound marketing/sales?
|
| I've seen multiple companies spin up outbound email
| marketing campaigns where someone compiles a list of 5000
| email addresses based on certain demographics, and then
| send automated emails (that look not automated) over the
| course of a month, rinse, repeat. Google Workspace will
| let you do this, but if you're too aggressive with email
| volume it can kill the reputation, and therefore
| deliverability of any email from that domain.
|
| (Which is why most companies send outbound sales emails
| from a domain other than their primary domain to separate
| out the sending domain reputation)
| 8organicbits wrote:
| > Does your company do outbound marketing/sales?
|
| Good guess, but we don't. I also checked DKIM/SPF when
| this happened and all appeared in order.
| derefr wrote:
| Plenty of dumps of stolen personal Gmail
| usernames+passwords, that anyone can feed into a bot that
| will use browser automation to sign into Gmail on those
| accounts and "hand write" some spam messages to send.
|
| (If you haven't realized, this is why Gmail has SMTP
| message origination disabled by default -- these days
| requiring not only enabling it for your Gmail account,
| but also fiddling with app passwords to get it working.
| If it was enabled by default, the "spam from stolen
| credentials" problem would be so, so much worse. Whereas,
| at least with the webapp route, Google can block you if
| you look like a bot [i.e. if you're doing an
| insufficiently good job at fooling them.])
| dalke wrote:
| 50% of my spam are from a single Google Groups source -
| unsolicited job applications, all ending:
| You received this message because you are subscribed to the
| Google Groups "jan-09" group. To unsubscribe from
| this group and stop receiving emails from it, send an
| email to jan-09+unsubscribe@googlegroups.com.
| To view this discussion on the web visit
| https://groups.google.com/d/msgid/jan-09/0
|
| 14101da368d$0bdc8160$23958420$@gmail.com.
|
| I cannot email that unsubscribe link because it says I am
| not subscribed. I cannot visit that page, I have not
| subscribed to that group. I've had to set up a special
| filter to look for that footer.
|
| I am not the only one with this issue. See
| https://support.google.com/groups/thread/68075070/i-get-
| goog... .
|
| ... Wait! You've indirectly helped solve the issue!
|
| They are being sent to "info@" my domain, an alias that
| forwards to my real account. I set up a new outgoing
| account with that From address, sent from there, and
| managed to get Google to unsubscribe something I never
| agreed to in the first place.
|
| It's been like this for a year, and with multiple attempts
| to fix it.
|
| Thank you!
| tempestn wrote:
| This reminds me, it would be nice if Google had an easy
| way to blacklist a sender and/or subject such that it
| wouldn't even go to spam. I have enough spam false
| positives that I needed to scan my spam folder
| periodically, and I'd love to have a way to permanently
| filter out regular crap in there. I've created explicit
| filters for some of the more prevalent ones, but a one
| click blacklist button on spam emails would be great.
| (Along with some way to edit the blacklist in case of
| mistakes.)
| iamacyborg wrote:
| There is a significant amount of spam coming from google
| accounts, yes. Just think about all the "sales automation"
| junk that businesses uses.
| cbsmith wrote:
| Because of course, its an arms race.
| op00to wrote:
| I switched from GMail to a personal Microsoft 365 domain when
| Google decided they didn't want to give me free email/domain
| services anymore. 365 was cheaper. I got about 10x the amount
| of spam to my 365 Junk folder than I did to the Junk folder in
| GMail. I would spend 10 minutes a day going through the junk
| folder to pick out false positives. I woud have inexplicable
| issues with missing email with 365, where the root cause was
| always SPF issues from a third party sender. The big issue was
| event tickets mailed from a third party ticket service provider
| using the venue's domain name rather than the ticket provider's
| domain.
|
| I switched back to GMail a few months ago, and not only do I
| see less stuff in my Junk folder (indicating Google is blocking
| stuff rather than identifying it) but also I have not seen a
| single false positive. Hopefully that means Google is more
| effective, but there's no way to tell if I'm missing legitimate
| email. So far, no complaints.
| yabones wrote:
| Microsoft's spam filter is fundamentally broken. It's been
| that way for decades. There's an entire cottage industry of
| snakeoil salesmen that want to sell in-line antispam gateways
| to bolt onto 365, and the worst part is that they have a very
| good reason to exist...
| JohnFen wrote:
| I've kinda given up on reputation scores to indicate spam/ham,
| personally, and rely more heavily on textual analysis rules.
| Going by "reputation" caused me far too many false positives.
| alexis2b wrote:
| > As an aside, not sure if OP is related to them but updown.io
| is a nice service and I appreciate the simple PAYG pricing! For
| what it's worth their mails seem to get through successfully to
| me too.
|
| Not related in any way except as an happy customer. They added
| a blog recently and this article caught my eye because of the
| nightmare that is mail delivery issue for everyone.
|
| I found it particularly ironic that you now have to think like
| a spammer (i.e. look at spam detection engine source code to
| find a way to circumvent their heuristics) in order to get your
| totally valid email delivered ( _^_^_ ).
|
| edit: typo
| jmyeet wrote:
| > ... the truth is that sender reputation is by far the biggest
| indicator of whether a message will be spam
|
| I couldn't agree with this more. I want people to remember this
| whenever the topic of decentralization or federation comes up.
| People see this as a technical problem. it's not. It's a
| political and organizational problem. Even with email, which is
| fully decentralized (other than the ICANN TLDs) running your
| own node still incredibly difficult. And those reasons aren't
| technical at all.
| matthews2 wrote:
| Putting your outbound emails through SpamAssassin as part of a
| regression test sounds like a really good idea - would have never
| thought of doing that myself!
| uean wrote:
| I love the analysis. But I hate that the 'fixed' email ends up
| being wordier for no reason at all.
|
| Brevity has value. Having to bloat content (an email to get past
| anti-spam; a cooking blog to rank better within Google SEO; ...)
| brings back memories of high-school english papers, or the modern
| equivalent ChatGPT.
| adrienjarthon wrote:
| 100% agree, I also hate that I had to do this.
| layer8 wrote:
| Couldn't you add some "hidden" text instead, e.g. white on
| white or display:none?
| adrienjarthon wrote:
| I could but I don't want to, it's even more of a dark
| pattern and looks way too "spammish" IMO. I don't want my
| users to find this in their email and think that I'm trying
| to trick their system. Also I wouldn't be suprised if some
| antispam tries to detect this as a spam criteria.
| chrismorgan wrote:
| Another piece of feedback: the link doesn't look like a
| _link_ any more. It wasn't great before, but the verbiage
| made it adequately clear. But now it's terrible, because the
| wording doesn't suggest an _action_ , and it doesn't look
| like a link _or_ a button. You should either restore its
| underline and lean into "link", or give a background colour
| or (generally better) gradient and lean into "button". But
| when it's just a border, it doesn't look like a button,
| especially when there's a tick after it. And change the
| wording again.
| adrienjarthon wrote:
| Thanks for this feedback, I actually changed this because
| some of my clients complained of the opposite, that the
| link was a bit too "dim" and didn't look like the the
| obvious Call To Action in the email. But it's all very
| debatable I agree and I may change this again in the
| future.
| chrismorgan wrote:
| The link style wasn't great before, just darker black
| with a faint underline. For best results, links should be
| underlined blue.
| jwr wrote:
| I've been using SpamAssassin since, well, forever, in internet
| terms. My recent facepalm moment was when I noticed that E-mails
| from the Playdate developer forum (Playdate is a really cool tiny
| gaming console) land in my spam folder, because anything in the
| .date domain (and the forum uses play.date as the domain) is
| assumed to be "dating spam".
| layer8 wrote:
| Given the double meaning of "play date", it's not surprising
| that it would cause a higher score, even if it used a different
| TLD.
| loloquwowndueo wrote:
| I have deny listed most of those new funny tld as it's indeed a
| good indication of spam. Here the face palm should be
| playdate's because they have realized their domain looks like a
| spam domain.
| JohnFen wrote:
| Yes, I do the same. It's very useful to me because there is
| no non-generic TLD that I would be getting legitimate email
| from, but it may not work well for people who do want to get
| emails from such TLDs.
| linsomniac wrote:
| It's been a very long time since I ran a mail server, but for a
| decade or more I pumped all our outgoing mail through Hashcash
| because it gave a good boost to the Spam Assassin score. We'd
| crank it through the largest one, and it would add ~60sec to the
| mail delivery, unless we had a bunch of outgoing mail, but it was
| worth it I felt.
| creeble wrote:
| I tried using SpamAssassin (via Proxmox Mail Gateway, which makes
| it much easier to set up) to replace a Barracuda email appliance
| (it was destined to get a *6x* service price increase in 2024!),
| and after several months of trying to get the number of FPs down,
| I gave up.
|
| The problem wasn't just the number of FPs (which were much higher
| than the 'Cuda) -- it was that they came from real people, who
| were often common senders. This is not corporate email, or
| anything that was even remotely spam (except as SA's crazy
| ruleset determined). These all required whitelisting, and it
| became a real chore for all my users to keep up with all the
| whitelisting.
|
| So back to the Barracuda for another year. It lets a little more
| spam through, but virtually no FPs. I just couldn't make SA get
| the same performance, even with many tweaks to the weights and
| rulesets.
| bschne wrote:
| I couldn't help but think about mechanistic interpretability
| research on large neural models reading this -- I guess this is
| what happens when humans do something similar, adding and
| removing tweaks here and there to better fit this or that case,
| over a long period of time.
| jaimex2 wrote:
| I won the war a while ago now.
|
| I basically trash all emails not in my contact lists. Easy.
___________________________________________________________________
(page generated 2024-01-02 23:01 UTC)