[HN Gopher] Google announces Sec-Gemini v1 a new experimental cy...
       ___________________________________________________________________
        
       Google announces Sec-Gemini v1 a new experimental cybersecurity
       model
        
       Author : ebursztein
       Score  : 86 points
       Date   : 2025-04-04 19:35 UTC (3 hours ago)
        
 (HTM) web link (security.googleblog.com)
 (TXT) w3m dump (security.googleblog.com)
        
       | esafak wrote:
       | It's interesting how we're seeing the emergence of specialized
       | models, much like trained humans.
        
         | jgalt212 wrote:
         | What's old is new again. Pretty much all ML and statistical
         | models were specialized for a single task / domain.
        
       | jruohonen wrote:
       | > Next, in response to a question about the vulnerabilities in
       | the Salt Typhoon description, Sec-Gemini v1 outputs not only
       | vulnerability details (thanks to its integration with OSV data,
       | the open-source vulnerabilities database operated by Google), but
       | also contextualizes the vulnerabilities with respect to threat
       | actors (using Mandiant data).
       | 
       | I remain still skeptical about LLMs in this space, although I
       | might be proven wrong, as often happens. Nevertheless, OSV has
       | already been a big advance, so it is great that it gets a further
       | commitment.
        
       | ziddoap wrote:
       | Could be great for _augmenting_ a cybersec professional 's tasks;
       | I'm certainly interested in trying it. However, I fear it will
       | not be used as just one of the tools in the toolbox, and rather
       | it will be used as something to defer (and consequently shed
       | liability) to.
        
         | booi wrote:
         | Has anybody been able to shed liability to AI yet?
        
           | ziddoap wrote:
           | In the legal sense? I'm not sure.
           | 
           | In the corporate day-to-day? Absolutely.
        
       | andy99 wrote:
       | Is this a "model" as in a set of transformer weights that
       | inherently does security work or is it a system that has data
       | lookup and or other tools along with an LLM to do the question
       | interpretation, synthesis, and output presentation?
       | 
       | From the description re data integrations it sounds like the
       | latter, unless the data mentioned is in fact used for training.
       | 
       | The distinction is important because a security-tuned model will
       | have different limitations and uses than an actual pre-build
       | security LLM app. Being an app also makes benchmarking against
       | other "models" less straightforward.
        
       | qwertox wrote:
       | There is generally something about the Gemini models which feels
       | a bit different than Claude, ChatGPT or Mistral.
       | 
       | I always have the feeling that I'm chatting with a model oriented
       | towards engineering tasks. The seriousness, lack of interest of
       | being humorous or cool.
       | 
       | I don't know if this is because I interact with Gemini only
       | through AI Studio, and it may have different system instructions
       | (apart from those one can add oneself, which I never do) than the
       | one at gemini.google.com.
       | 
       | I never use gemini.google.com because of the lack of a simple
       | export feature. And it's not even possible to save one chat to
       | disk (well, neither do the others), I just wish it did.
       | 
       | AI Studio saving to Google Drive is really useful. I lets you
       | download the chat, strip it of verbose things like the thinking
       | process, and reuse it in a new chat.
       | 
       | I wish gemini.google.com had a "Save as Markdown" per answer and
       | for the complete chat (with a toggle to include/exclude the
       | thinking process). Then it would be a no brainer for me.
       | 
       | It's the same as if Google Docs would not have an "Download.."
       | menu entry but you could only "save" the documents via Takeout.
        
         | Y_Y wrote:
         | > The seriousness, lack of interest of being humorous or cool.
         | 
         | I love this. When ChatGPT compliments me on my great question
         | or tries to banter it causes me great despair.
        
           | neodypsis wrote:
           | I've noticed 4o uses a lot of emojis, and, in general, is
           | very enthusiastic. I find it funny. If I want a more formal
           | bot, I switch to one of the o3 family.
        
             | jonplackett wrote:
             | Every now and then 4o seems to get a bit drunk and use
             | tonnes of emojis or start swearing when I haven't sworn
             | myself in the chat.
             | 
             | The other day I asked a fairly innocuous question and it
             | LOLed and said it'd give me the 'no Bullshit answer'
        
               | kfajdsl wrote:
               | I've had 4o start off its response with a Smiling Face
               | with Sunglasses emoji by the heading unprompted lol.
               | 
               | edit: does hacker news filter out emojis? TIL (there
               | should be emojis after this colon: )
        
             | querez wrote:
             | I use a very simple custom system prompt (not on my work
             | machine at the moment, but essentially something along the
             | lines of "for technical questions, please be concise and to
             | the point, and when asked for code, omit explanations and
             | emit just the code itself unless I ask for explanations"),
             | and it does wonders.
        
           | qwertox wrote:
           | So do I. But it's not like ChatGPT isn't flexible, the code
           | it generates for small tasks is really good, and the site is
           | faster than AI Studio.
           | 
           | For example, if I want to quickly create a Python script to
           | list all VMs via libvirt and output their attached drives and
           | filesystems, that's a task for ChatGPT.
           | 
           | But for the things where I don't want an AI to "suck up" to
           | me and instead "stay professional", that's Gemini.
        
         | HelenePhisher wrote:
         | > And it's not even possible to save one chat to disk (well,
         | neither do the others), I just wish it did.
         | 
         | Ask Claude to generate a .md of the conversation, it will do
         | that with the option to download that or a PDF of it. A lovely,
         | but well hidden feature!
        
           | qwertox wrote:
           | Thanks for the tip. I tested it and this also works with
           | Gemini and ChatGPT.
           | 
           | The only drawback I see is that it requires enough free space
           | in the context window to duplicate the visual part of the
           | chat.
        
         | gavinray wrote:
         | It doesn't seem as popular, but I've found Grok to treat you
         | the least like a child and provide good answers. Especially
         | with more complicated tasks.
        
           | uejfiweun wrote:
           | I think Grok is the best for asking about current events but
           | I kind of hate how it always tries to turn everything into a
           | conversation. But that's just my opinion! What do you think
           | is the most annoying feature about Grok?
           | 
           | ^ like that.
        
         | asadm wrote:
         | 2.5 has been amazing for programming. I just send it entire
         | repo as context when I am lazy and then ask it for entire
         | modified files back with the (medium sized) change. It almost
         | always works! I wish to either start using cursor or some
         | vscode extension to do this from ide itself.
        
         | occamschainsaw wrote:
         | I have been using the Obsidian web clipper to export chats from
         | ChatGPT and Claude web versions to nicely-formatted md files.
         | You can save md to Obsidian or download it as a standalone
         | file. It doesn't support Gemini yet though.
         | 
         | https://github.com/obsidianmd/obsidian-clipper
        
         | tyushk wrote:
         | You put into words something I've been struggling to describe
         | for a long time. Gemini gives short, succinct responses with
         | whatever information you need and minimal anything else.
         | ChatGPT, Claude both fill text with mannerisms, formatting,
         | etc.
         | 
         | I didn't realize just how big the difference was until I tested
         | it.
         | 
         | "How do I clear a directory of all executable files on Debian?"
         | 
         | Gemini 2.0 Flash: (responses manually formatted)
         | find /path/to/directory -type f -executable -delete
         | Replace /path/to/directory with the actual path.
         | 
         | ChatGPT: (full link [1])                   To clear (delete)
         | all executable files from a directory on Debian (or any Linux
         | system), you can use the find command. Here's a safe and
         | effective way to do it:         # [checkmark emoji] Command to
         | delete all executable files in a directory (not recursively):
         | [..]         # [magnifying glass emoji] Want to preview before
         | deleting? [..]         # [caution sign emoji] Caution: [..]
         | 
         | [1]
         | https://chatgpt.com/share/67f055c8-4cc0-8003-85a6-bc1c7eadcc...
        
       ___________________________________________________________________
       (page generated 2025-04-04 23:00 UTC)