hlfshell

@hlfshell@hachyderm.io

https://hlfshell.ai

👤 A person who likes to make stuff #maker #diy 🛠️

📚 An avid reader with an ever-growing to-read list #books #bookrecs 📙

💾 Software Engineer specializing #robotics , distributed data and #deeplearning training pipelines, and backend infrastructure ⚙️

🤖 Currently studying for a Masters in Robotics Engineering @ #WPI 🦾

🏖️🌊🌴
Located in San Diego, CA
🏖️🌊🌴

This profile is from a federated server and may be incomplete. Browse more on the original instance.

hlfshell, to random

My favorite interview question to ask, from the perspective of the hiring manager, is:

"What do you wish I would have asked you about? A technical question, something on your resume, anything. Just talk to me for five minutes about something you wish you could talk about in an interview."

You'll see some people light up, excited, and learn something you'd never find in any defined process. You'll see them dive into the details of their successes, or tell you about a hobby project, or more.

hlfshell, to llm

Read through this paper last night https://arxiv.org/abs/2402.04494
TLDR Deepmind repurposed a standard LLM to accept a singular tokenized gameboard (not game sequence) and produce an action-value function (probability of winning off of the move == score) out of the transformers in it. With no additional game search they achieved grandmaster equivalence; with two caveats.

hlfshell, to ai

What if we looked at large language models with neuroscience techniques instead of taking a mechanical view?

That question was the center of the Representation Engineering paper, and I covered it here. Learn about control vectors, how easy it is to use them, and how you can gain powerful control over your language models.

https://hlfshell.ai/posts/representation-engineering/

#ai #llm #neuroscience

hlfshell,

@kellogh I actually think they might, because it provides amazing warning triggers and jailbreak protection.

And yeah, I do think there will be a lot of great tooling built around these ideas soon.

kellogh, to random
@kellogh@hachyderm.io avatar

in 2012 most tech people were worried about a bubble. Also in 2013-15, and 2016-20, and 2021-now. i thought for sure the crypto crash was going to trigger a larger bubble pop. yet somehow, it still hasn't happened.

not saying it won't happen, but there's a whole lot of history behind naysayers being wrong

hlfshell,

@kellogh The general rule for dealing with absurd stock pricing and bubbles, especially when betting against it, is that the market can stay irrational longer than you can stay solvent.

hlfshell, to llm

Has anyone tried combining traditional symbolic AI techniques with LLM models to offload logic/reasoning?

I'm talking about more than just considering RAG + search as a knowledge graph. Basically, papers that discuss things like LLM self generation and modification of knowledge graphs.

#llm #ai

kellogh, to python
@kellogh@hachyderm.io avatar

imo #python isn’t a good scripting language bc environment config is difficult enough that you can’t often write scripts that reliably run when called

hlfshell,

@kellogh Depends on how people define scripting language which can be a loosely thrown around term.

Personally whenever I want actually reliable tooling I just fall back to Go, since 99% of the time you can just create a portable executable with no environment management needed after that.

luis_in_brief, to random
@luis_in_brief@social.coop avatar

Really enjoying the Arc browser. The only painful thing is with each new fun, innovative feature, thinking how much of this Firefox could have been doing a decade ago.

hlfshell,

@luis_in_brief open source could very much benefit from a product designer. Case in point as well is Chrome OS - there was nothing about Chrome OS Linux and Firefox and others couldn't have tried to make - a simple isolated browser only OS.

And again with Slack. What if IRC but user friendly? Hell that was literally how it was built. And then it became Discord. All things that could have been done in open source but it fell short due to user friendliness and design being a different skill set.

hlfshell,

@luis_in_brief is it chromium based of a new engine built from the ground up?

hlfshell,

@luis_in_brief makes sense to build off of chromium but there's serious concerns about locking the web down to just one target engine (two I guess due to Safari). Really want to see innovation in the tough realm of the browser engine too.

BlackAzizAnansi, to random
@BlackAzizAnansi@mas.to avatar

The older I get the more to the left i get pushed and I was told that the opposite would happen lol.

hlfshell,

@BlackAzizAnansi It was a myth. There's been a few studies on it over a long period of time. The general rule is that political beliefs are very stable throughout the lifetime of most people, but when there is a shift it tends to be towards more liberal beliefs.

This might be an exasperated outcome, however, as certain parties have shifted so far right that having stable conservative beliefs in comparison makes the individual seem like they're becoming more liberal.

SteveThompson, to random
@SteveThompson@mastodon.social avatar

"Elon Musk’s worst nightmare"

https://www.businessinsider.com/missy-cummings-robot-taxis-tesla-elon-musk-autonomous-cruise-waymo-2023-10

"Cummings...is more than happy to strafe companies like Tesla and Waymo and Cruise, and...argue that tech bros need to be brought inside a stricter regulatory framework. In a sense, she's Elon Musk's worst nightmare. She has repeatedly and routinely risked her life to test...capabilities — and...lethal limits — of human-machine interfaces. To her, the safety of self-driving cars is not an abstract question. It's a matter of life and death."

hlfshell,

@mastodonmigration @SteveThompson anything that has a penalty of purely a fine is simply legalizing it for a high enough economic class.

GossiTheDog, to random
@GossiTheDog@cyberplace.social avatar

Everest ransomware group want to cut out middleman fees, asking to buy corporate access directly.

hlfshell,

@GossiTheDog This makes perfect sense. One disgruntled or underpaid employee is all it takes to get their in. I'm surprised I haven't seen other groups so openly advertise it.

hlfshell, to llm

Just finished reading Promptbreeder. An interesting idea oddly explained and executed.

tldr; if you can define a fitness function for prompts you can use LLMs to mutate and crossover a set of prompts to slowly evolve better performing ones.

https://arxiv.org/abs/2309.16797

#llm #ai #promptengineering

hlfshell, to random

Every single piece of research about work from home, vacation policies, employee pay, or vacation time essentially reads as

"Hey if you treat your employees like human beings and not a cog in a machine to chase quarterly numbers you get really productive employees and low turnover."

And every single business is like "All these empirical studies are cool and all but this one 50+ year old white guy VP FEELS differently so OH WELL..."

Followed by shocked Pikachu face when dealing with turnover.

baldur, to random
@baldur@toot.cafe avatar

I expected the AI bubble to get bad, but the scale of the bullshit still hits hard

Only those who weren’t likely to get involved in it anyway seem to have heeded any warnings

Everything I see in the LLM space is close to the diametric opposite of what I would have recommended

hlfshell,

@baldur any particularly egregious examples to share?

hlfshell,

@baldur not great. It seems people latch onto the most obvious use of LLMs, mass text content generation, when it's also the least useful due to their problems.

Then again textbooks are shit quality in the US if your state has a certain ideology. This is just that exacerbated.

Khan Academy had a better approach, using LLMs to generate interactive problem discussion.

But those interactive applications are harder to build right and therefore not the quick to market product.

hlfshell, to llm

Extremely suspicious that all the super cool multi agent coding AIs that supposedly can build complete applications are always doing the same demos, consisting of popular tutorial sample apps.

Can anyone link me to more impressive demos or write ups of performance? #llm #ai

GossiTheDog, to Starfield
@GossiTheDog@cyberplace.social avatar

deleted_by_author

    •
    hlfshell,

    @GossiTheDog If it lives up to about 70% of the hype I'll pick it up. I don't have high hopes of it meeting that expectation, but I'll be a very happy space-camper if it does.

    AJCxZ0, to random

    While I stopped being a young adult around the time I purchased glasses in order to be able to see small things up close, I am the perfect audience member for @pluralistic's brilliant "Little Brother".

    Half "Nineteen Eight-Four", half introduction to privacy and cryptography, half hacker manifesto, half teen romance, half historical account, one quarter nostalgia hit and five quarters thrilling adventure well read by Kirby Heyborne.

    https://craphound.com/category/littlebrother/

    #LittleBrother #CoryDoctrow

    hlfshell,

    @AJCxZ0 @pluralistic I enjoyed the series. I read all three last year (not sure if the third one is considered spin off or a legit part of the series). Each book had a strong theme:

    1. Introduce a techies version of why unlimited surveillance is terrible, how easily abused it is.

    2. Focus on the profiteering side of the military industrial surveillance complex.

    3. How can grass roots organizations hope to overcome the legions of well funded spy industry goons?

    The first one was my fav.

    MissingThePt, to random
    @MissingThePt@mastodon.social avatar

    "Twitter/X plans to disable the block feature, impacting both user experience and safety.”

    "There's this alternative called Mastod-"

    "And it’s beholden to one vain, insecure, capricious man obsessed with fighting Zuck."

    "There's this federated option Mast-"

    "Not to mention it refuses to suspend pro-Hitler accounts and unsuspends CSAM accounts because it’s lucrative."

    "The non-profit Ma-"

    "Welp, nothing to be done about Twitter/X I suppose, might as well stick it out on this hellhole."

    hlfshell,

    @MissingThePt but how can I comfortably talk about stuff without some billionaire making money off of me? 🤑 💰 🔥

    hlfshell,

    @IamDohnut @ticho @MissingThePt generally follow hashtags. Take an
    few minutes, search for related hashtags, and follow the tags themselves.

    Then follow people that use the tags over time as you find if they might be interesting.

    andypiper, to random
    @andypiper@macaw.social avatar

    In a meeting today, someone said to me: "... you reaaaally don't like Musk, do you?!".

    No. No I do not.

    hlfshell,

    @andypiper He could have just faded into obscurity with infinite-generational wealth, but he is insistent on forcing himself into the public eye for as much attention as possible...

    ....while making things terrible for a lot of people.

    He deserves all the hate he's getting.

    GossiTheDog, to random
    @GossiTheDog@cyberplace.social avatar

    deleted_by_author

    •
    hlfshell,

    @GossiTheDog Why would you do this to us

    f, to tv

    Something I've noticed about a lot of American TV shows — the first season starts off with a great idea but by season two or three it has become all about the personal lives of the characters and their relationships and the core idea gets relegated to the background and gets lost.

    “House” was a great medical drama at first and I loved it for the medicine but eventually it became all drama and not much medicine. The same with “The Good Doctor”.

    (I can contrast this to UK TV but I guess that’ll have to be a post for another time 😛)

    “Warrior” (https://www.imdb.com/title/tt5743796) is a show which doesn’t do this as much and which manages to maintain a balance between the core story and the character development. Or maybe, it’s because the story is all about the characters 🙂

    But what I do like about the show a lot is the fact that the characters learn, grow, and change. Characters who started out looking as if they are pure evil start getting shades of grey as you learn about them and their motivations. Others change their attitudes and stance as time goes on.

    The show is constantly changing and alliances constantly shift. And it’s marvellous to see these happening and to see a story being told so deftly 🙂

    As the end of season 3 approaches, I can see a lot of changes and a bunch of new alliances happening. I hope there’s a season 4 since I want to keep on watching!

    hlfshell,

    @stevesplace @f I'm hard pressed to stay nemesis; I have a hard time being too mad at artists just trying to earn a paycheck.

    ....especially when the likely incentive is almost certainly due to studio executives.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • JUstTest
  • mdbf
  • ngwrru68w68
  • tester
  • magazineikmin
  • thenastyranch
  • rosin
  • khanakhh
  • InstantRegret
  • Youngstown
  • slotface
  • Durango
  • kavyap
  • DreamBathrooms
  • megavids
  • tacticalgear
  • osvaldo12
  • normalnudes
  • cubers
  • cisconetworking
  • everett
  • GTA5RPClips
  • ethstaker
  • Leos
  • provamag3
  • anitta
  • modclub
  • lostlight
  • All magazines
    •